Home Docs Claude Developer Platform 9 Nov 2026

One change · api

optimizing-for-cost-and-intelligence changedabout-claude/models/optimizing-for-cost-and-intelligence

Nearest release: v2.1.321, published under an hour before this site recorded the change. Shown because the two are within 24 hours of each other. Nothing here says the release caused the edit.

Recorded here 9 Nov 2026 19:07 UTC

Lines+85added

Lines−83removed

From line 23 where the diff opens

First seen 14 Aug 2026 this site's first read of the page

Recorded edits36to this page, all time

The whole hunk

from line 23, old and new numbered

lines

from line 23

2323

2424Match your situation to a row.

2525

26| Your situation | Do this | Where |

27| ------------------------------------------------ | ---------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |

28| Any workload, any model | Turn on prompt caching and trim unneeded tokens; both are free | [Cache repeated context](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#cache-repeated-context) · [Trim tokens](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#trim-input-and-context-tokens) |

29| A person waits between turns | Count the gaps between your requests, then pick the cache duration from the table in that section | [Pick the cache duration](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#pick-the-cache-duration) |

30| Costs are too high; quality is fine | Try lower effort on the model you use now | [Tune effort](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#tune-effort) |

31| You are not on the latest model | Upgrade; in Anthropic’s measurements each newer model solved at least as many tasks as the one before it, usually for less per solved task | [Upgrade the model](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#upgrade-the-model) |

32| You are choosing or switching models | Compare on cost per completed task, not per token | [Compare models](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#compare-models-on-cost-per-task) |

33| High-volume work with outputs you can check | Try Claude Haiku 5.5 first | [Compare models](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#compare-models-on-cost-per-task) |

34| Quality isn’t good enough | If you lowered effort, restore it; otherwise try the next tier up at its default effort | [Tune effort](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#tune-effort) · [Compare models](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#compare-models-on-cost-per-task) |

35| Attempts end with `stop_reason: max_tokens` | Raise `max_tokens` to 64,000: in Anthropic’s tests only 2 of about 14,000 turns were longer, and cost per solved task stayed about the same | [Set budgets](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#set-budgets-and-output-caps) |

36| You can check outputs (tests, a verifier) | Run at `low` effort and re-run failures at `high`: in Anthropic’s tests, about 40% lower cost than everything at `high`, at the same pass rate | [Re-run failures](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#re-run-failures-at-higher-effort) |

37| Agent loops with a few very costly runs | Set a Claude Managed Agents session budget and a workspace spend limit | [Set budgets](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#set-budgets-and-output-caps) |

38| You want agent runs to finish sooner | Tell the model time matters and show it the elapsed time: 7% to 47% less time in Anthropic’s tests, for a few points of score | [Show the model elapsed time](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#show-the-model-elapsed-time) |

39| A lower-cost model stalls only on hard decisions | Try an advisor. It can help only when your model consults it; first price its model alone at low effort and measure the consult rate | [Advisor strategy](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#advisor-strategy-escalate-hard-decisions) |

40| The work exceeds one context window | Split the work and hand the parts to lower-cost models | [Orchestrator strategy](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#orchestrator-strategy-delegate-bulk-work) |

26| Your situation | Do this | Where |

27| ------------------------------------------------ | ------------------------------------------------------------------------------------------------------------------------------------------------------- | ----------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------------- |

31| You are not on the latest model | Upgrade; on the coding benchmarks in that section each newer model solved at least as many tasks as the one before it, usually for less per solved task | [Upgrade the model](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#upgrade-the-model) |

35| Attempts end with `stop_reason: max_tokens` | Raise `max_tokens` to 64,000: in Anthropic’s tests only 7 of about 18,000 turns were longer, and cost per solved task did not rise | [Set budgets](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#set-budgets-and-output-caps) |

4141

4242These results are Anthropic-internal ([Benchmarks referenced](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs)) and directional, not guarantees, so measure on your own workload with the [four-step method](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#measure-on-your-own-workload).

4343

from line 52

5252Turn on [prompt caching](https://platform.claude.com/docs/en/build-with-claude/prompt-caching) before any other lever, because by default every turn of an agentic task resends the entire growing conversation: system prompt, tool definitions, and every prior turn.

5353

5454* A 40-turn task sends its first turn 40 times, so task cost grows with roughly the square of turn count.

55* Caching does not stop the resending, but each resend costs about a tenth as much (a twentieth on Claude Opus 5.5 and Claude Sonnet 5.5, a fortieth on Claude Fable 5.1 and Claude Mythos 5.1) and processes faster.

55* Caching does not stop the resending, but each resend costs about a tenth as much (a twentieth on Claude Opus 5.5 and Claude Sonnet 5.5, a fortieth on Claude Fable 5.2, Claude Mythos 5.2, Claude Fable 5.1, and Claude Mythos 5.1) and processes faster.

5656* The prefix is billed at the [cache-read rate](https://platform.claude.com/docs/en/build-with-claude/prompt-caching#pricing) instead of the input price, and each turn pays the 1.25x cache-write rate only for what is new.

5757

5858[Message threads](https://platform.claude.com/docs/en/build-with-claude/message-threads), in beta, stop the resending while a thread lasts, until 5 minutes after its last request. They lower time to first token, not cost: the model still reads the whole conversation, so keep `cache_control` on every request. Threads don’t support compaction or most context editing.

from line 90

9090

9191#### Pick the cache duration

9292

93If your loop waits on a person between turns, use the [1-hour cache duration](https://platform.claude.com/docs/en/build-with-claude/prompt-caching#1-hour-cache-duration) once about 1 turn in 20 follows a pause between 5 minutes and an hour and few gaps run over an hour. On Claude Fable 5.1, Claude Opus 5.5, and Claude Sonnet 5.5, keeping the 5-minute cache warm can cost less.

93If your loop waits on a person between turns, use the [1-hour cache duration](https://platform.claude.com/docs/en/build-with-claude/prompt-caching#1-hour-cache-duration) once about 1 turn in 20 follows a pause between 5 minutes and an hour and few gaps run over an hour. On Claude Fable 5.2, Claude Fable 5.1, Claude Opus 5.5, and Claude Sonnet 5.5, keeping the 5-minute cache warm can cost less.

9494

9595The 1-hour duration costs more to write (2x the input price instead of 1.25x). A miss on either duration bills the whole prefix at the write price instead of the read price.

9696

from line 107

107107| Gaps over an hour are common: about 60% or more of your pauses longer than 5 minutes also run past an hour | Any | 5-minute default |

108108| Only 1 or 2 gaps in 20 fall between 5 minutes and an hour, and none lasts more than about half an hour | Claude Opus 5.5 and Claude Sonnet 5.5 | 5-minute cache kept warm, instead of the 1-hour duration |

109109| A pause comes before every turn | Claude Opus 5.5 and Claude Sonnet 5.5 | 1-hour duration |

110| Pauses run for minutes | Claude Fable 5.1 | 5-minute cache kept warm, instead of the 1-hour duration |

111| Pauses run toward an hour | Claude Fable 5.1 | 1-hour duration |

110| Pauses run for minutes | Claude Fable 5.1 and Claude Fable 5.2 | 5-minute cache kept warm, instead of the 1-hour duration |

111| Pauses run toward an hour | Claude Fable 5.1 and Claude Fable 5.2 | 1-hour duration |

112112

113113On Claude Haiku 5.5, use the “Any” rows: it reads the cache at the standard 0.1x the input price, twice the 0.05x of Claude Opus 5.5 and Claude Sonnet 5.5, so by list prices each keep-alive request costs twice as much, relative to the input price, as on those two models. A `max_tokens: 0` request refreshed its cache in a separate test (reference 16).

114114

from line 125

125125 * **Only 1 or 2 gaps in 20.** At default effort, keep-alive requests cost 8% to 13% less than the 1-hour duration on Claude Opus 5.5, 8% to 16% less on Claude Sonnet 5.5.

126126 * **A pause comes before every turn.** On Claude Opus 5.5 and Claude Sonnet 5.5, keep-alive requests cost a few percent more than the 1-hour duration at 6-minute pauses and over 50% more at 45-minute pauses.

127127

128* **Claude Fable 5.1**, whose [cache reads](https://platform.claude.com/docs/en/about-claude/pricing#prompt-caching) cost 0.025x the input price.

128* **Claude Fable 5.1 and Claude Fable 5.2**, whose [cache reads](https://platform.claude.com/docs/en/about-claude/pricing#prompt-caching) cost 0.025x the input price.

129129

130 * **Pauses run for minutes.** On Claude Fable 5.1, keep-alive requests[19](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) cost 13% to 20% less than the 1-hour cache, with 10% of turns paused and with a 6-minute pause before every turn.

131 * **Pauses run toward an hour.** On Claude Fable 5.1, only with a 45-minute pause before every turn did the 1-hour cache win, by about $0.12 USD on a session of 5 bug reports.

130 * **Pauses run for minutes.** On Claude Fable 5.2, keep-alive requests[19](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) cost about 4.5% to 19% less than the 1-hour cache, with 10% of turns paused and with a 6-minute pause before every turn.

131 * **Pauses run toward an hour.** On Claude Fable 5.2, only with a 45-minute pause before every turn did the 1-hour cache win, by about $0.12 USD[19](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) on a session of 5 bug reports.

132132

133133The following chart plots cost per session against the share of paused turns on Claude Sonnet 5.5:

134134

from line 145

145145**What to know.**

146146

147147* The 1-in-20 rule leaves a margin over the break-even of about 1 turn in 30. At 1 turn in 20 the 5-minute and the 1-hour setting cost about the same on Claude Sonnet 5.5, within run-to-run variation; at 1 turn in 10 the 5-minute setting cost 54% more (reference 16).

148* On models other than Claude Sonnet 5.5 and Claude Opus 5.5, prices put the break-even in the same range: every current model uses the same cache-write multipliers, and every model but Claude Fable 5.1, Claude Mythos 5.1, Claude Opus 5.5, and Claude Sonnet 5.5 the same read price.

148* On models other than Claude Sonnet 5.5 and Claude Opus 5.5, prices put the break-even in the same range: every current model uses the same cache-write multipliers, and every model but Claude Fable 5.2, Claude Mythos 5.2, Claude Fable 5.1, Claude Mythos 5.1, Claude Opus 5.5, and Claude Sonnet 5.5 the same read price.

149149* The Claude Opus 5.5 and Claude Sonnet 5.5 keep-alive figures come from requests sent with `max_tokens: 1`. A `max_tokens: 0` request also refreshed an existing cache entry on Claude Fable 5.1, Claude Opus 5.5, Claude Sonnet 5.5, and Claude Haiku 5.5 (reference 16).

150150

151151To keep the cache warm:

from line 158

158158

159159A `max_tokens: 0` request is rejected in the following cases:

160160

161* The request sets `thinking.type: "enabled"` (the default adaptive thinking on Claude Fable 5.1 is fine), structured outputs, or a forced tool choice ([its limitations](https://platform.claude.com/docs/en/build-with-claude/prompt-caching#limitations)). On those workloads, buy the 1-hour duration instead.

161* The request sets `thinking.type: "enabled"` (the default adaptive thinking on Claude Fable 5.1 and Claude Fable 5.2 is fine), structured outputs, or a forced tool choice ([its limitations](https://platform.claude.com/docs/en/build-with-claude/prompt-caching#limitations)). On those workloads, buy the 1-hour duration instead.

162162* The request carries the top-level `compaction` parameter, so don’t re-send a compaction request from [compaction on demand](https://platform.claude.com/docs/en/build-with-claude/compaction-on-demand) as a keep-alive request.

163163

164164<CodeGroup exclude="shell:CLI, python, typescript, csharp, go, java, php, ruby">

from line 198

198198 ![Bar chart: system prompt edit up to $1.26 USD, top-level effort change up to $1.02 USD, either in a message up to $0.026 USD](https://platform.claude.com/docs/images/cost-intel/cache-break-cost.svg)

199199</Frame>

200200

201At list prices, the stakes are highest on Claude Fable 5.1 and Claude Mythos 5.1, where a break rewrites the prefix at 1.25x the input price instead of reading it at 0.025x. On Claude Haiku 5.5, a system prompt edit cost 12.5x as much as an unchanged request, because the edit rewrote the whole conversation. On models that run the [prefix check](https://platform.claude.com/docs/en/build-with-claude/preserved-thinking#prefix-check), a request that makes this edit and sends earlier thinking blocks back can get a 400 error instead[18](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs). The top-level effort change, in a conversation that held the model’s first reply with its thinking, rewrote all but the system prompt. These figures are from the Messages API.

201At list prices, the stakes are highest on Claude Fable 5.2, Claude Mythos 5.2, Claude Fable 5.1, and Claude Mythos 5.1, where a break rewrites the prefix at 1.25x the input price instead of reading it at 0.025x. On Claude Haiku 5.5, a system prompt edit cost 12.5x as much as an unchanged request, because the edit rewrote the whole conversation. On models that run the [prefix check](https://platform.claude.com/docs/en/build-with-claude/preserved-thinking#prefix-check), a request that makes this edit and sends earlier thinking blocks back can get a 400 error instead[18](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs). On Claude Fable 5.2 and Claude Mythos 5.2, every account gets the 400 error by default, including accounts created before August 31, 2026. The top-level effort change, in a conversation that held the model’s first reply with its thinking, rewrote all but the system prompt. These figures are from the Messages API.

202202

203203**What to do.**

204204

from line 274

274274* Run-wide, the prune and context editing cache about equally well. If you use context editing, [clear in a few large batches](https://platform.claude.com/docs/en/build-with-claude/context-editing#context-editing-and-prompt-caching).

275275

276276<Note>

277 On Claude 5.1 and later models, rewriting an earlier tool result [invalidates every thinking block after it](https://platform.claude.com/docs/en/build-with-claude/preserved-thinking#prefix-check), so use compaction instead. The prune was measured on an earlier model without [preserved thinking](https://platform.claude.com/docs/en/build-with-claude/preserved-thinking); on these models, [context editing](https://platform.claude.com/docs/en/build-with-claude/context-editing) and [compaction](https://platform.claude.com/docs/en/build-with-claude/compaction) run on the server and don’t invalidate thinking blocks. [Send assistant turns back exactly as returned](https://platform.claude.com/docs/en/build-with-claude/preserved-thinking#append-assistant-turns-exactly-as-returned): dropping their thinking blocks costs output tokens on every turn, because Claude [re-creates the reasoning](https://platform.claude.com/docs/en/build-with-claude/preserved-thinking#drop-block-cost).

277 On Claude Fable 5.2, Claude Mythos 5.2, Claude Fable 5.1, Claude Mythos 5.1, Claude Opus 5.5, Claude Sonnet 5.5, and Claude Haiku 5.5, rewriting an earlier tool result [invalidates every thinking block after it](https://platform.claude.com/docs/en/build-with-claude/preserved-thinking#prefix-check), so use compaction instead. The prune was measured on an earlier model without [preserved thinking](https://platform.claude.com/docs/en/build-with-claude/preserved-thinking); on these models, [context editing](https://platform.claude.com/docs/en/build-with-claude/context-editing) and [compaction](https://platform.claude.com/docs/en/build-with-claude/compaction) run on the server and don’t invalidate thinking blocks. [Send assistant turns back exactly as returned](https://platform.claude.com/docs/en/build-with-claude/preserved-thinking#append-assistant-turns-exactly-as-returned): dropping their thinking blocks costs output tokens on every turn, because Claude [re-creates the reasoning](https://platform.claude.com/docs/en/build-with-claude/preserved-thinking#drop-block-cost).

278278</Note>

279279

280280The prune, adapted from the triage agent’s code:

from line 384

384384

385385## Trade cost against intelligence

386386

387These levers set where a single model sits between cost and intelligence: model choice, effort, re-running failures at a higher setting, the budgets and caps it works within, and whether it can see how much time has passed. If you are a model or two behind, upgrade first ([Upgrade the model](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#upgrade-the-model)); otherwise, start with an effort sweep on your current model ([Tune effort](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#tune-effort)). From lowest to highest cost and capability, the current models are Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus 5.5, and Claude Fable 5.1 (the frontier model); [Models overview](https://platform.claude.com/docs/en/models/overview) has the full lineup and prices.

387These levers set where a single model sits between cost and intelligence: model choice, effort, re-running failures at a higher setting, the budgets and caps it works within, and whether it can see how much time has passed. If you are a model or two behind, upgrade first ([Upgrade the model](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#upgrade-the-model)); otherwise, start with an effort sweep on your current model ([Tune effort](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#tune-effort)). From lowest to highest cost and capability, the current models are Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus 5.5, and Claude Fable 5.2 (the frontier model); [Models overview](https://platform.claude.com/docs/en/models/overview) has the full lineup and prices.

388388

389389### Compare models on cost per task

390390

from line 405

409409| GPQA Diamond[9](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) | Claude Haiku 5.5 | `medium` (default) | 85% accuracy, compared with 91% for each of Claude Opus 5.5 and Claude Sonnet 5.5 | About a twentieth of Claude Opus 5.5’s cost per correct answer and about a tenth of Claude Sonnet 5.5’s |

411411

412On 463 of the 478 problems in the SWE-bench Pro subset, the ones whose reference solutions also pass with no network access (reference 3), Claude Fable 5.2 at `high`, its default on the Messages API, solved 96.3% of tasks for $1.39 USD per solved task. The same Claude Fable 5.1 run at its default, counted over these 463 problems, solved 92.4% for $1.17 USD. At `low`, Fable 5.2 solved 94.6% for $0.39 USD per solved task: more tasks than Fable 5.1 at its default, for about a third of the cost. At a similar pass rate, Claude Opus 5.5 still costs less per solved task on these problems: at its default it solved 93.1% for $0.22 USD, at `high` 95.5% for $0.30 USD, and at `xhigh` 96.5% for $0.75 USD, level with Fable 5.2 at `high` for a little over half the cost.

413

412414**What to do.**

413415

414* Start with Claude Opus 5.5 at its default effort: on the SWE-bench Pro subset it matched Claude Fable 5.1 at its default, inside run-to-run noise, for about a fifth of the cost per solved task. Opus 5.5 and Fable 5.1 both largely saturate the SWE-bench Pro subset. Use Claude Fable 5.1 for demanding reasoning and long-horizon agentic work, or when your evals on Claude Opus 5.5 at higher effort still fall short.

416* Start with Claude Opus 5.5 at its default effort: on the SWE-bench Pro subset it matched Claude Fable 5.1 at its default, inside run-to-run noise, for about a fifth of the cost per solved task. Opus 5.5 and Fable 5.1 both largely saturate the SWE-bench Pro subset. Use Claude Fable 5.2 for demanding reasoning and long-horizon agentic work, or when your evals on Claude Opus 5.5 at higher effort still fall short.

415417* At the other end, Claude Haiku 5.5 suits high-volume and latency-sensitive work with checkable outputs.

416418* Price every candidate in cost per completed task on your own traffic, including Claude Opus 5.5 at its default effort and the frontier model at reduced effort. Each of Claude Opus 5.5’s token prices, cache reads included, is twice Claude Sonnet 5.5’s, but on the SWE-bench Pro subset Opus 5.5 at `low` solved about the same share of tasks as Sonnet 5.5 at its default, inside run-to-run noise, for about 7% more per solved task.

417* Measure what a more expensive model buys on your own workload. On the 45 DeepResearch Bench II[7](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) tasks scored for every model, at default effort, Claude Opus 5.5 scored 2.6 points above Claude Sonnet 5.5 for 1.5 times the cost per task, and Claude Fable 5.1 scored 6.0 points above Claude Sonnet 5.5 for 5.8 times. On all 50 tasks, with four biology tasks that only Claude Sonnet 5.5 answered counted as failures, neither is measurably ahead of Claude Sonnet 5.5.

419* Measure what a more expensive model buys on your own workload. On the 44 DeepResearch Bench II[7](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) tasks scored for every model, at default effort on the Messages API, Claude Opus 5.5 scored 2.4 points above Claude Sonnet 5.5 for 1.5 times the cost per task. Claude Fable 5.2 cost 4.3 times as much as Claude Sonnet 5.5 for no measurable gain in score, and 2.9 times as much as Claude Opus 5.5 for a score 1.9 points lower. Against Claude Fable 5.1, it cost 20% to 61% less at each effort level and scored 4.7 to 7.0 points lower. On all 50 tasks, Claude Opus 5.5 is not measurably ahead of Claude Sonnet 5.5: safeguards nearly always stopped it on four biology tasks that Claude Sonnet 5.5 answered, and a stopped attempt counts as a failure.

418420* Price the tail of your workload, not the median: compare models on the hardest tenth of your tasks, not the typical one.

419421

420422The bill is decided by the tasks the cheaper model fails, because a failed task still bills its tokens, then the retry, then whatever the failure costs downstream. The tail is also where the money goes even when nothing fails. On a 20-problem WideSearch[1](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) run, two problems carried 43% of the spend:

from line 429

427429

428430### Upgrade the model

429431

430If you are a model or two behind, the cheapest lever is the model string. In Anthropic’s measurements, each newer model solved at least as many tasks as the one before it, usually for less per solved task. The chart plots recent Claude models on two benchmarks in which an agent works alone on one written task and tests check the result: the SWE-bench Pro[3](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) subset, where every model ran through the same test setup, and Terminal-Bench 3[20](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs). The table lists what each upgrade changed:

432If you are a model or two behind, the cheapest lever is the model string. On the coding benchmarks in this section and in [Compare models](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#compare-models-on-cost-per-task), each newer model solved at least as many tasks as the one before it; most cost less per solved task, and Claude Fable 5.2, at `high`, cost more than Claude Fable 5.1. On a research benchmark, Claude Fable 5.2 did the reverse: it cost less per task than Claude Fable 5.1 at each effort level and scored lower ([Compare models](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#compare-models-on-cost-per-task)). The chart plots recent Claude models on two benchmarks in which an agent works alone on one written task and tests check the result: the SWE-bench Pro[3](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) subset, where every model ran through the same test setup, and Terminal-Bench 3[20](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs). The table lists what each upgrade changed:

431433

432434<Frame>

433435 ![Charts: on SWE-bench Pro, Opus 5.5 and Sonnet 5.5 cut cost per solved task most; on Terminal-Bench 3, Opus drops to $28 USD](https://platform.claude.com/docs/images/cost-intel/upgrade-ladder-sonnet-5-5.svg)

from line 454

452454

453455### Tune effort

454456

455Sweep effort on your own workload before you add a second model. Lower effort costs less and is often faster, and how much accuracy it gives up depends on the task: on long-horizon coding, effort bought accuracy, and on research tasks it bought a lot in one measurement and a little in another. The [`effort`](https://platform.claude.com/docs/en/build-with-claude/effort) parameter governs how much thinking, tool calling, and self-verification the model does; `high` is the default on most models, and `medium` on Claude Opus 5.5 and Claude Haiku 5.5.

457Sweep effort on your own workload before you add a second model. Lower effort costs less and is often faster, and how much accuracy it gives up depends on the task: on long-horizon coding, effort bought accuracy, and on research tasks it bought a lot in one measurement and a little in another. The [`effort`](https://platform.claude.com/docs/en/build-with-claude/effort) parameter governs how much thinking, tool calling, and self-verification the model does; on the Messages API, `high` is the default on most models, and `medium` on Claude Opus 5.5 and Claude Haiku 5.5.

456458

457459The following table lists Anthropic’s measurements on two benchmarks with current models; scores and costs are not comparable between the two benchmarks. Costs are per task, not per solved task as in [Compare models on cost per task](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#compare-models-on-cost-per-task). Each score pools two runs on SWE-bench Pro and three on DeepResearch Bench II, unless its cell says otherwise:

458460

460| -------------------------------------------------------------------------------------------------------------------------------------- | ---------------- | ------------------ | ------------------------------------ | ------------- |

461| SWE-bench Pro[3](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) subset | Claude Opus 5.5 | `low` | 87.4% of tasks solved | $0.11 USD |

465| DeepResearch Bench II[7](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) | Claude Fable 5.1 | `low` | Rubric score 60.5% | $3.47 USD |

462| -------------------------------------------------------------------------------------------------------------------------------------- | ---------------- | ------------------------------------ | ------------------------------------ | ------------- |

463| SWE-bench Pro[3](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) subset | Claude Opus 5.5 | `low` | 87.4% of tasks solved | $0.11 USD |

467| DeepResearch Bench II[7](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) | Claude Fable 5.2 | `low` | Rubric score 54.5% | $1.49 USD |

468470

469471The following chart plots accuracy against cost per task for Claude Opus 5.5 on the SWE-bench Pro[3](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) subset:

470472

from line 485

483485* Sweep two or three effort levels on a sample of your own traffic, each in a separate session: changing top-level effort mid-session [invalidates the cache](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#cache-repeated-context) and distorts the comparison. The curve is the baseline any multi-model strategy must beat ([step 2 of measuring on your own workload](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#measure-on-your-own-workload)).

484486* On long-horizon coding, effort is a real tradeoff, which [re-running failures at higher effort](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#re-run-failures-at-higher-effort) turns back into a saving. At `medium`, Claude Sonnet 5.5 scored about 7 points lower than at its default, `high`, on the SWE-bench Pro[3](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) subset for about 59% of the cost per task (2 runs each), and scored 62.7% compared with 77.8% on the internal agentic-coding benchmark[11](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) (1 attempt per task).

485487* At `xhigh` and `max`, Claude Sonnet 5.5 can start its own rounds of review after it finishes a task, which takes more time and tokens. To keep that work on the task, add the prompt under “Thoroughness at `xhigh` and `max` effort” in [Prompting Claude Sonnet 5.5](https://platform.claude.com/docs/en/build-with-claude/prompt-engineering/prompting-claude-sonnet-5-5#steer-initiative-and-scope) to your system prompt.

486* On research tasks, measure the model you ship. Claude Fable 5.1 scored 2.5 points higher at `high` than at `low` on DeepResearch Bench II, for 1.6 times the cost, and no higher at `medium` than at `low`; Claude Opus 5.5 scored 13.3 points lower at `low` than at `medium` on DRACO[21](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs). The two results come from different models on different benchmarks, so they show that the answer varies, not why.

488* On research tasks, measure the model you ship. Claude Fable 5.2 scored 3.3 points higher at `high` than at `low` on DeepResearch Bench II, for 3.2 times the cost, and not measurably higher at `medium` than at `low`; Claude Opus 5.5 scored 13.3 points lower at `low` than at `medium` on DRACO[21](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs). The two results come from different models on different benchmarks, so they show that the answer varies, not why.

487489* At `low`, Claude Opus 5.5’s typical task took 84% less time than at `medium` on DRACO[21](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs), 30% less on HLE[22](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs), and 47% less on an internal set of physics problems[23](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs), and it scored 13.3, 3.4, and 14.0 points lower (3, 3, and 4 attempts per task).

488* On an internal defect-search benchmark[8](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) (10 runs per setting), Claude Opus 5.5 scored 34 points lower at `low` than at `medium`, for about a fifth of the cost. Claude Fable 5.1 scored 1.8 points lower at `low` than at `high`, for about a quarter less.

490* On an internal defect-search benchmark[8](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs) (10 runs per setting), Claude Opus 5.5 scored 34 points lower at `low` than at `medium`, for about a fifth of the cost. Claude Fable 5.2 scored 4.3 points lower at `low` than at `high`, for 37% less.

489491

490492### Re-run failures at higher effort

491493

from line 522

520522* Set `max_tokens` to 64,000 for agentic work, or to 128,000, the maximum, when a single cut-off attempt is costly. [Stream responses](https://platform.claude.com/docs/en/build-with-claude/streaming) that large, treat [`stop_reason: max_tokens`](https://platform.claude.com/docs/en/build-with-claude/handling-stop-reasons#max-tokens) as a failure, and save money with effort, which the model can see.

521523* On Claude Haiku 5.5, raise a small `max_tokens` or lower effort: thinking is on by default and counts toward the cap, so a response can end before any text (see the [Claude Haiku 5.5 migration guide](https://platform.claude.com/docs/en/models/haiku-5-5/migration-guide#configure-thinking)).

522524

523The following table shows what each model solved and what a solved task cost at each cap on an internal repository-task benchmark[12](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs), one attempt per task, each model at its default effort. The rows at 16,384 average two runs, and the other rows are one run each:

525The following table shows what each model solved and what a solved task cost at each cap on an internal repository-task benchmark[12](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs), one attempt per task, each model at its default effort on the Messages API. The rows at 16,384 average two runs, and the other rows are one run each:

524526

526| ---------------- | ------------ | ------------------------- | ------------ | ------------------------ |

527| Claude Opus 5.5 | 16,384 | 25% | 57.6% | $5.38 USD at list prices |

528| Claude Opus 5.5 | 64,000 | 0% | 75.6% | $5.38 USD at list prices |

529| Claude Opus 5.5 | 128,000 | 0% | 68.9% | $5.94 USD at list prices |

530| Claude Fable 5.1 | 16,384 | 43% | 36.3% | $21.37 USD as billed |

531| Claude Fable 5.1 | 64,000 | 0% | 58.5% | $21.53 USD as billed |

532| Claude Fable 5.1 | 128,000 | 1% | 60.0% | $21.74 USD as billed |

528| ---------------- | ------------ | ------------------------- | ------------ | ------------------------- |

529| Claude Opus 5.5 | 16,384 | 25% | 57.6% | $5.38 USD at list prices |

530| Claude Opus 5.5 | 64,000 | 0% | 75.6% | $5.38 USD at list prices |

531| Claude Opus 5.5 | 128,000 | 0% | 68.9% | $5.94 USD at list prices |

532| Claude Fable 5.2 | 16,384 | 85% | 20.3% | $27.96 USD at list prices |

533| Claude Fable 5.2 | 64,000 | 2% | 77.4% | $24.42 USD at list prices |

534| Claude Fable 5.2 | 128,000 | 0% | 78.2% | $24.52 USD at list prices |

533535

534536**Claude Opus 5.5 at 64,000 and 128,000:** No attempt was cut off at either cap, so the cap does not explain why these two rows differ in tasks solved or in cost per solved task. Each row is one run.

535537

536Compare the caps within each model. The two models were run a month apart, and their costs are priced on different bases.

538Compare the caps within each model. The two models were measured in separate runs.

537539

538On the internal repository tasks[12](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs), almost every turn finishes far below a 16,384-token cap. The rare long turn is what a higher cap buys:

540On the internal repository tasks[12](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs), almost every turn finishes far below a 16,384-token cap, but one longer turn ends the whole attempt: with a higher cap, 1.7% of Claude Fable 5.2’s turns passed 16,384 tokens, and at 16,384 the cap ended 85% of its attempts. The rare long turn is what a higher cap buys (Claude Opus 5.5 and Claude Fable 5.1 are the two models in this chart):

539541

540542<Frame>

541543 ![Dot plot of per-turn output for Opus 5.5 and Fable 5.1: medians a few hundred tokens, longest 61k and 128k, against the caps](https://platform.claude.com/docs/images/cost-intel/max-tokens-ladder-opus-5-5.svg)

542544</Frame>

543545

544Ask for shorter answers. Output tokens cost five times input tokens on Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus 5.5, and Claude Fable 5.1, and in an agent loop every token the model writes comes back as input on every later turn, so you pay for a long answer again and again.

546Ask for shorter answers. Output tokens cost five times input tokens on Claude Haiku 5.5, Claude Sonnet 5.5, Claude Opus 5.5, Claude Fable 5.2, and Claude Fable 5.1, and in an agent loop every token the model writes comes back as input on every later turn, so you pay for a long answer again and again.

545547

546548In a measurement on an earlier model[26](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs), Anthropic ran a small agent that triages bug reports ([Trim input and context tokens](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#trim-input-and-context-tokens)) under three final-answer instructions, three runs each, with the same model and tools:

547549

from line 573

571573 ![Scatter and bar charts, DRACO, HLE, and the physics set: the changes save the most time on DRACO, for a few points of score](https://platform.claude.com/docs/images/cost-intel/time-awareness.svg)

572574</Frame>

573575

574To pay for speed instead, set `speed: "fast"` on Claude Opus 5.5. [Fast mode](https://platform.claude.com/docs/en/build-with-claude/fast-mode) gives up to 2.5x higher output tokens per second at twice the standard token prices. Switching speed mid-conversation misses the prompt cache.

576To pay for speed instead, set `speed: "fast"` on Claude Fable 5.2 or Claude Opus 5.5. [Fast mode](https://platform.claude.com/docs/en/build-with-claude/fast-mode) gives up to 2.5x higher output tokens per second at twice the standard token prices. Switching speed mid-conversation misses the prompt cache.

575577

576578## Combine models

577579

from line 640

638640 ![Diagram of the orchestrator strategy: a frontier model fans subtasks out to three lower-cost worker models](https://platform.claude.com/docs/images/model-routing-orchestrator-strategy.svg)

639641</Frame>

640642

641The following chart plots 9 of the defect-search benchmark’s 12 configurations by cost and score:

643The following chart plots 9 of the defect-search benchmark’s 17 configurations by cost and score (none of Claude Fable 5.2’s five):

642644

643645<Frame>

644646 ![Scatter chart, defect search: Opus 5.5 with Sonnet 5.5 subagents matches Opus 5.5 alone, at medium, for about a third less](https://platform.claude.com/docs/images/cost-intel/corpus-pareto.svg)

from line 649

647649**What to do.**

648650

649651* **Routine work with a long cost tail:** measure this case first, for accuracy as well as cost. The measurement behind it covers only problems that the delegated configuration answered correctly in every run (reference 4).

650* **Work larger than one context window:** split it into parts, one session or one subagent per part. On the defect-search benchmark[8](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs), Claude Opus 5.5 with Claude Sonnet 5.5 subagents cost about a third less than one Claude Opus 5.5 session per part at its default effort, and scored about the same. Claude Fable 5.1 as the orchestrator in Claude Opus 5.5’s place cost 24% more for no measurable gain. Alone, Claude Fable 5.1 cost 2.2 to 3.0 times as much as Claude Opus 5.5 at `high` and scored no higher.

652* **Work larger than one context window:** split it into parts, one session or one subagent per part. On the defect-search benchmark[8](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs), Claude Opus 5.5 with Claude Sonnet 5.5 subagents cost about a third less than one Claude Opus 5.5 session per part at its default effort, and scored about the same. Claude Fable 5.2 as the orchestrator in Claude Opus 5.5’s place cost 19% more and scored about the same. Alone, from `low` to `high`, Claude Fable 5.2 scored 1.6 to 5.9 points higher than Claude Opus 5.5 alone at `high`, for 1.5 to 2.4 times the cost, and 3.2 to 7.5 points higher than with Claude Sonnet 5.5 subagents, for 2.2 to 3.4 times the cost. It also scored higher than Claude Fable 5.1 alone at each effort level: 3.6 points at `low`, for 32% less; 3.8 at `medium`, for 37% less; and 6.1 at `high`, for 21% less.

651653* **Claude Haiku 5.5 subagents on coding tasks:** compare them with lower effort first. On an internal coding benchmark on Claude Managed Agents, they were an alternative to lowering effort, not better. Adding them to Claude Opus 5.5 at `medium`, with a short delegation prompt, cut cost per attempt by 32% and the pass rate by about 4 percentage points (95% interval 1.3 to 7.1), about what a mix of `low` and `medium` without subagents would give at that cost[29](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#refs).

652654

653655## Measure on your own workload

from line 866

864866| Data files through code execution | 92% on a data task of 25 questions, with prompt caching off | A gain: 25 of 25 correct instead of 7 | Faster | [Trim input and context tokens](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#trim-input-and-context-tokens) |

866868| Prompt audit against the current model | 14% on both model upgrades tested | None; a gain on one | Faster (fewer tool rounds) | [Audit prompts against the current model](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#audit-prompts-against-the-current-model) |

867| Upgrade the model | Usually less: up to 40% less on typical workloads, and up to 87% less per solved task on a coding benchmark | None; a gain on most | Neutral | [Upgrade the model](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#upgrade-the-model) |

868| Lower effort | On long coding tasks, `medium` cost about 30% to 52% less than `high`, and `low` about two thirds less | Up to 34 points lower, depending on the task and the model; 2.5 on one research agent | Faster | [Tune effort](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#tune-effort) |

869| Upgrade the model | Usually less: up to 40% less on typical workloads, and up to 87% less per solved task on a coding benchmark | None on coding, a gain on most; lower on one research benchmark for Claude Fable 5.2 | Neutral | [Upgrade the model](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#upgrade-the-model) |

870| Lower effort | On long coding tasks, `medium` cost about 30% to 52% less than `high`, and `low` about two thirds less | Up to 34 points lower, depending on the task and the model; 3.3 on one research agent | Faster | [Tune effort](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#tune-effort) |

869871| Re-run failures | About 40% less than running everything at `high`, at the same pass rate or slightly better | None | A second attempt on the tasks that fail | [Re-run failures at higher effort](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#re-run-failures-at-higher-effort) |

871| Raising `max_tokens` | None: cost per solved task stayed about the same | A gain: up to 22 points more tasks solved on one coding benchmark; none on another | Neutral | [Set budgets and output caps](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#set-budgets-and-output-caps) |

873| Raising `max_tokens` | None: cost per solved task did not rise at 64,000 | A gain: up to 57 points more tasks solved at 64,000 on one coding benchmark; none on another | Neutral | [Set budgets and output caps](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#set-budgets-and-output-caps) |

872874| Advisor | None: about 2.1 times the cost of the model alone, on a coding benchmark | On coding, a small gain that may be chance; on chart reading, a loss | About 1.4 extra calls per attempt, on coding | [Advisor strategy](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#advisor-strategy-escalate-hard-decisions) |

873875| Orchestrator | With subagents on the next model down, about a third less than the orchestrator’s model alone at its default effort, on work larger than one context window; on routine work, measure first | No measurable loss on that work | Up to 25 subagents at a time | [Orchestrator strategy](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#orchestrator-strategy-delegate-bulk-work) |

874876| Fast mode | None: twice the standard token prices | None | Up to 2.5x higher output tokens per second | [Show the model elapsed time](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#show-the-model-elapsed-time) |

from line 927

925927 * **Advisor, Claude Sonnet 5 pairings:** The Claude Sonnet 5 executor pairings with a Claude Opus 5 advisor behind the consult-rate examples in [Advisor strategy](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#advisor-strategy-escalate-hard-decisions) ran on this subset through the same test setup: the pairing at the default effort was run twice (August 7 and August 8, 2026, a run and an exact replication), the low-effort pairing once (August 8, 2026), and, ten days later, Claude Sonnet 5 alone twice at its default effort (77.4%, the baseline for both pairings).

926928 * **Advisor, Claude Sonnet 5.5 consult check:** The Claude Sonnet 5.5 consult check cited in [Advisor strategy](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#advisor-strategy-escalate-hard-decisions) ran one attempt on each of 20 of the 478 problems, on September 24, 2026, with a Claude Opus 5.5 advisor through the advisor tool. Four configurations ran: `high` and `xhigh`, each with and without the advisor tool documentation’s suggested system prompt for coding tasks. At its default effort, `high`, Claude Sonnet 5.5 did not consult the advisor on any of the 20 problems, with or without the prompt. At `xhigh`, it consulted on 3 of the 20 problems without the prompt and 4 with it, each time on its attempts with the most turns.

927929 * **Escalation figures:** Escalation figures come task by task from the Opus 5.5 runs: `low` first, then `high` on its failures, solved 96.4% to 97.5% across run pairings for about $0.17 USD; `medium` first, 96.0% to 97.1% for about $0.24 USD; `high` re-run on its own failures, 96.9% for $0.31 USD; everything at `high`, 94.8% to 95.8% for $0.29 USD.

928 * **Compare models:** The Claude Fable 5.1 figure in [Compare models](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#compare-models-on-cost-per-task) comes from one run on the same subset at the default effort, run August 26, 2026: counted over the 478 problems, it solved 92.3% for $1.19 USD per solved task. That section’s chart and its Claude Opus 5.5 and Claude Sonnet 5.5 figures pool the two runs at each setting plotted, all counted over the 478 problems: Claude Opus 5.5 at `low`, `medium`, and `high`, and Claude Sonnet 5.5 at `medium` and `high`.

930 * **Compare models:** The Claude Fable 5.1 figure in [Compare models](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#compare-models-on-cost-per-task) comes from one run on the same subset at the default effort, run August 26, 2026: counted over the 478 problems, it solved 92.3% for $1.19 USD per solved task. That section’s chart and its table’s Claude Opus 5.5 and Claude Sonnet 5.5 figures pool the two runs at each setting plotted, all counted over the 478 problems: Claude Opus 5.5 at `low`, `medium`, and `high`, and Claude Sonnet 5.5 at `medium` and `high`. Claude Fable 5.2 figures are one run each at `low` and at `high` (its default on the Messages API), both with the same 16,384-token cap per turn, which cut off 8 of the 463 `high` attempts and none at `low`; cut-off attempts count as failures. One more `high` attempt was stopped by a safety monitor in the evaluation environment before it finished and also counts as a failure, although the changes it had made by then passed the tests; leaving it out would give 96.5% for `high`. Those runs used a version of the benchmark whose grading containers have no network access. Of the 478 problems used for the Opus 5.5 comparisons, the reference solution fails without network access on 15, so the Fable 5.2 comparisons use the remaining 463. In those comparisons, the Claude Opus 5.5 figures are the same runs counted over the 463 problems: two runs each at `medium` and `high`, and one at `xhigh`. The Claude Fable 5.1 and Claude Opus 5.5 figures set beside them were graded with network access, so only the problem set is matched, not the grading setup. Claude Fable 5.2’s costs use the metering described under How priced, applied to each attempt’s token totals rather than to each request separately, so they are estimates.

929931 * **Compare models, work per task:** In those figures, Claude Opus 5.5 at `low` did less work on each task than Claude Sonnet 5.5 at its default: a little over half of Sonnet 5.5’s output tokens and a little over a third of its cache reads. At their defaults, Opus 5.5 at `medium` solved about 5 percentage points more than Sonnet 5.5 for about twice the cost. At that setting it did slightly more work per task than Sonnet 5.5.

930932 * **Upgrade the model, runs:** The Claude Fable 5 point in Upgrade the model pools three runs at the default effort, run August 26, 2026, priced as a customer’s organization is metered. The upgrade ladder counts every model over the 478 problems, each at its shipped defaults. It uses one run per model except Opus 5, Opus 5.5, Sonnet 5, and Sonnet 5.5 (two runs each) and Fable 5 (the three runs described above). The Opus 4.7, Opus 4.8, Sonnet 4.6, and Sonnet 5 runs ran August 17 to 19, 2026, and the Opus 5 runs on August 4, all in one harness and organization; the Opus 5.5 and Sonnet 5.5 runs are the ones described earlier in this reference.

931933 * **Upgrade the model, cost per solved task:** Claude Opus 5 costs 21% more per solved task than Claude Opus 4.8 when those Claude Opus 5 runs are compared with one Claude Opus 4.8 run at its shipped defaults from the August 17 to 19 series, counted over the same 478 problems; the chart’s SWE-bench Pro panel shows neither Claude Opus 4.7 nor Claude Opus 4.8. In the same measurements, priced as a customer is billed, Claude Opus 5 solves 12 more points of tasks than Claude Opus 4.8 at that 21% more per solved task, Claude Opus 4.8 solves the same share of tasks as Claude Opus 4.7 for 14% less per solved task, and Claude Opus 5 at `low` effort beats Claude Opus 4.8’s default on this benchmark for about 30% of its cost per solved task.

from line 943

941943

9429447. **DeepResearch Bench II:** Li et al., “DeepResearch Bench II: Diagnosing Deep Research Agents via Rubrics from Expert Report,” arXiv:2601.08536, 2026.

943945

944 * **What was measured:** Its 132 research tasks across 22 domains are graded against expert-derived binary rubrics. Measured October 22, 2026, on a 50-task subset drawn from all its themes, half in English and half in Chinese, one attempt per task, 3 runs per setting: Claude Sonnet 5.5 at `high`, Claude Opus 5.5 at `medium`, and Claude Fable 5.1 at `low`, `medium`, and `high`. Every setting ran on the Messages API with adaptive thinking and the production safety classifiers on.

945 * **Tools:** The agent used search and fetch tools that the evaluation harness hosts over an Anthropic web index with nothing dated after March 17, 2026, and had no shell. A search query over 120 characters was cut at the last whole word, and the agent was told so: 48% of searches on Claude Sonnet 5.5, and 26% to 31% in the other settings. About a quarter of fetches returned an error. Each task’s source article and the benchmark’s own paper and dataset were filtered out.

946 * **How scored:** Claude Opus 4.6 judges under the benchmark’s rubric protocol; the original uses a different judge, and an Anthropic judge may favor the house style. Scores are means over the 45 tasks that every setting answered and the judge fully graded. The other five: four biology tasks that the safety classifiers refused in every run of Claude Opus 5.5 and Claude Fable 5.1 and in none of Claude Sonnet 5.5’s, and one on which a safety check that Anthropic adds to its own evaluations stopped an attempt.

947 * **Scores:** On the 45 tasks, and on all 50, with a refused or stopped attempt counted as 0: Claude Sonnet 5.5, 56.9% and 57.0%; Claude Opus 5.5, 59.5% and 54.3%; Claude Fable 5.1, 60.5% and 55.1% at `low`, 60.5% and 54.9% at `medium`, and 63.0% and 57.4% at `high`. Task-by-task differences on the 45 tasks, taken before rounding, with 95% intervals: Claude Fable 5.1 at `high` over `low`, 2.5 points (1.3 to 3.6); Claude Opus 5.5 over Claude Sonnet 5.5, 2.6 (0.7 to 4.5); Claude Fable 5.1 at `high` over Claude Opus 5.5, 3.4 (2.0 to 4.9), and over Claude Sonnet 5.5, 6.0 (4.4 to 7.7); Claude Fable 5.1 at `medium` over `low`, 0.0 (-1.3 to 1.2). On all 50 tasks: Claude Opus 5.5 over Claude Sonnet 5.5, -2.7 (-8.3 to 2.9); Claude Fable 5.1 at `high` over Claude Sonnet 5.5, 0.4 (-5.3 to 6.2).

948 * **How priced:** Each cost per task is the model’s own tokens at list prices, from the attempts’ usage records, over all 50 tasks and every attempt, averaged over the three runs: $1.05 USD on Claude Sonnet 5.5, $1.52 USD on Claude Opus 5.5, and $3.47 USD, $4.44 USD, and $5.64 USD on Claude Fable 5.1. A refused attempt costs about a cent. On the 45 tasks alone the costs are $1.06 USD, $1.63 USD, $3.75 USD, $4.81 USD, and $6.10 USD. The cost ratios in [Compare models](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#compare-models-on-cost-per-task), 1.5 and 5.8, are on those 45 tasks, like the scores beside them; over all 50 tasks they are 1.4 and 5.4. Costs cover model tokens only; the harness’s tools add no charges. The agent made 17 to 28 searches per task, refused attempts counted.

949 * **Caching chart:** The caching chart re-prices the same requests with every input token at the uncached rate. The evaluation harness set the cache breakpoints.

946 * **What was measured:** Its 132 research tasks across 22 domains are graded against expert-derived binary rubrics. Measured November 3, 2026, on a 50-task subset drawn from all its themes, half in English and half in Chinese, one attempt per task, 3 runs per setting: Claude Sonnet 5.5 at `high`, Claude Opus 5.5 at `medium`, and Claude Fable 5.1 and Claude Fable 5.2 at `low`, `medium`, and `high`. Every setting ran on the Messages API with adaptive thinking, and the three released models with the production safety classifiers on. Claude Fable 5.2 was run before its release, on 46 of the 50 tasks, without the four biology tasks.

947 * **Tools:** The agent used search and fetch tools that the evaluation harness hosts over an Anthropic web index with nothing dated after March 17, 2026, and had no shell. A search query over 120 characters was cut at the last whole word, and the agent was told so: 57% to 59% of searches on Claude Fable 5.2, 49% on Claude Sonnet 5.5, and 26% to 31% in the other settings. Within a task, Claude Fable 5.2’s runs with more cut searches did not score lower. Further runs of Claude Fable 5.2 and Claude Opus 5.5 followed on November 6 and 7, 2026, three per model in each of two sets. The tool cut longer queries as before in one set and refused them in the other. On the 43 tasks scored in both sets, Claude Fable 5.2 at `high` scored 0.9 points below Claude Opus 5.5 with the cut (95% interval 1.0 above to 2.7 below). That interval includes the 1.9-point gap measured on November 3, on 44 tasks. With refusal it scored 1.9 points below (0.0 to 3.7). That gap was 1.0 points larger than with the cut (0.8 smaller to 2.8 larger), not a measurable change. About a quarter of fetches returned an error. Each task’s source article and the benchmark’s own paper and dataset were filtered out.

948 * **How scored:** Claude Opus 4.6 judges under the benchmark’s rubric protocol; the original uses a different judge, and an Anthropic judge may favor the house style. Scores are means over the 44 tasks that every setting answered and the judge fully graded. The other six: four biology tasks that Claude Sonnet 5.5 answered in every run, and on which the safety classifiers stopped every attempt of Claude Fable 5.1 (36) and 11 of Claude Opus 5.5’s 12; one task, a survey of computer security research, on which the classifiers stopped one attempt of Claude Fable 5.1 at `low` and two of Claude Fable 5.2’s three at `high`; and one on which a safety check that Anthropic adds to its own evaluations stopped one attempt of Claude Sonnet 5.5.

949 * **Scores:** On the 44 tasks, and on all 50, with a refused or stopped attempt counted as 0 and one answered attempt that the judge did not fully grade kept at its partial score: Claude Sonnet 5.5, 57.2% and 56.8%; Claude Opus 5.5, 59.6% and 54.6%; Claude Fable 5.1, 60.2% and 54.7% at `low`, 62.1% and 56.6% at `medium`, and 62.5% and 57.0% at `high`. Claude Fable 5.2, on the 44 tasks: 54.5% at `low`, 55.0% at `medium`, and 57.7% at `high`. Task-by-task differences on the 44 tasks, taken before rounding, with 95% intervals: Claude Fable 5.1 at `high` over `low`, 2.3 points (1.3 to 3.3); Claude Opus 5.5 over Claude Sonnet 5.5, 2.4 (1.0 to 3.8); Claude Fable 5.1 at `high` over Claude Opus 5.5, 2.8 (1.4 to 4.3), and over Claude Sonnet 5.5, 5.2 points (4.0 to 6.5); Claude Fable 5.1 at `medium` over `low`, 1.9 (0.6 to 3.1); Claude Fable 5.2 at `high` over `low`, 3.3 (1.7 to 4.9), and at `medium` over `low`, 0.6 (-0.6 to 1.8); Claude Fable 5.2 at `high` over Claude Sonnet 5.5, 0.5 (-1.2 to 2.2), and over Claude Opus 5.5, -1.9 (-3.6 to -0.2); Claude Fable 5.2 over Claude Fable 5.1, -5.8 (-8.0 to -3.5) at `low`, -7.0 (-9.3 to -4.8) at `medium`, and -4.7 (-6.4 to -3.1) at `high`. On all 50 tasks: Claude Opus 5.5 over Claude Sonnet 5.5, -2.2 (-7.6 to 3.2); Claude Fable 5.1 at `high` over Claude Sonnet 5.5, 0.2 (-5.5 to 5.9).

950 * **How priced:** Each cost per task is the model’s own tokens at list prices, from the attempts’ usage records, over all 50 tasks and every attempt, averaged over the three runs: $1.11 USD on Claude Sonnet 5.5, $1.61 USD on Claude Opus 5.5, and $3.55 USD, $4.44 USD, and $5.69 USD on Claude Fable 5.1. Claude Fable 5.2’s costs are over its 46 tasks: $1.49 USD, $2.16 USD, and $4.82 USD; over the same 46 tasks the other five settings cost $1.13 USD, $1.69 USD, $3.85 USD, $4.82 USD, and $6.18 USD. On the four biology tasks, each Claude Fable 5.1 attempt was stopped at its first request, and attempts cost about a cent on average on Claude Fable 5.1 and $0.74 USD on Claude Opus 5.5. On the 44 tasks alone the costs are $1.13 USD, $1.68 USD, $3.85 USD, $4.77 USD, and $6.10 USD, and $1.50 USD, $2.20 USD, and $4.86 USD on Claude Fable 5.2. The cost comparisons in [Compare models](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#compare-models-on-cost-per-task) (1.5, 4.3, and 2.9 times; 61%, 54%, and 20% less than Claude Fable 5.1 at `low`, `medium`, and `high`) and in [Tune effort](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#tune-effort) (3.2 times) are on those 44 tasks, like the scores beside them. Costs cover model tokens only; the harness’s tools add no charges. The agent made 12 to 33 searches per task, refused attempts counted.

951 * **Charts:** The caching chart and the effort chart show runs from October 22, 2026, of Claude Fable 5.1, Claude Opus 5.5, and Claude Sonnet 5.5 at the same effort levels. The caching chart re-prices those runs’ requests with every input token at the uncached rate. The evaluation harness set the cache breakpoints.

950952 * **Comparability:** These scores are not comparable with the benchmark’s leaderboard.

951953

9529548. **Internal defect-search benchmark:**

953955

954 * **What was measured:** Anthropic-internal, for work larger than one context window: a corpus of about 22 million tokens from 14 public Python package sources, in 201 parts, with 130 planted defects and deterministic grading. Ten runs per configuration, on [Claude Managed Agents](https://platform.claude.com/docs/en/managed-agents/overview): 51 of the 120 on October 27 to 29, 2026, and 69 on November 2 to 3, 2026.

955 * **Configurations:** Alone, each model ran one session per part, each session given the findings so far: Claude Fable 5.1 and Claude Opus 5.5 at `low`, `medium`, and `high`, and Claude Sonnet 5.5 and Claude Haiku 5.5 at each model’s default (`high` and `medium`). With subagents, Claude Fable 5.1 or Claude Opus 5.5 ran at `high` as the orchestrator, through [multiagent orchestration](https://platform.claude.com/docs/en/managed-agents/multiagent-orchestration) (the `multiagent_20261001` type): one predefined agent as the subagent, at its model’s default effort, called once for each part, up to 25 at a time, with inline agents and dynamic workflows turned off. Every configuration ended with one more session, without subagents, that assembled the findings. Thinking ran at the platform’s default: on, with the model choosing when to think.

956 * **What was measured:** Anthropic-internal, for work larger than one context window: a corpus of about 22 million tokens from 14 public Python package sources, in 201 parts, with 130 planted defects and deterministic grading. Ten runs per configuration, on [Claude Managed Agents](https://platform.claude.com/docs/en/managed-agents/overview): 51 of the 170 on October 27 to 29, 2026, 69 on November 2 to 3, 2026, and Claude Fable 5.2’s 50 on November 3 to 4, 2026, before its release.

957 * **Configurations:** Alone, each model ran one session per part, each session given the findings so far: Claude Fable 5.2, Claude Fable 5.1, and Claude Opus 5.5 at `low`, `medium`, and `high`, and Claude Sonnet 5.5 and Claude Haiku 5.5 at each model’s default (`high` and `medium`). With subagents, Claude Fable 5.2, Claude Fable 5.1, or Claude Opus 5.5 ran at `high` as the orchestrator, through [multiagent orchestration](https://platform.claude.com/docs/en/managed-agents/multiagent-orchestration) (the `multiagent_20261001` type): one predefined agent as the subagent, at its model’s default effort, called once for each part, up to 25 at a time, with inline agents and dynamic workflows turned off. Every configuration ended with one more session, without subagents, that assembled the findings. Thinking ran at the platform’s default: on, with the model choosing when to think.

956958 * **Tools:** Every agent used the benchmark’s own shell and file tools, in an isolated virtual machine. Before any agent started, the evaluation removed installed copies of the corpus’s packages from that machine and checked that no connection from it succeeded.

957 * **Results:** Mean F1 (range) and mean cost per run. Claude Haiku 5.5 alone: 0.577 (0.550 to 0.606), $25 USD. Claude Sonnet 5.5 alone: 0.760 (0.669 to 0.788), $103 USD. Claude Opus 5.5 alone at `low`: 0.489 (0.437 to 0.551), $38 USD; at `medium`: 0.827 (0.797 to 0.850), $200 USD; at `high`: 0.850 (0.829 to 0.876), $236 USD. Claude Fable 5.1 alone at `low`: 0.829 (0.807 to 0.841), $527 USD; at `medium`: 0.839 (0.817 to 0.856), $603 USD; at `high`: 0.847 (0.829 to 0.861), $716 USD. Claude Opus 5.5 with Claude Sonnet 5.5 subagents: 0.831 (0.814 to 0.850), $139 USD; with Claude Haiku 5.5 subagents: 0.677 (0.640 to 0.718), $47 USD. Claude Fable 5.1 with Claude Sonnet 5.5 subagents: 0.825 (0.812 to 0.858), $172 USD; with Claude Haiku 5.5 subagents: 0.687 (0.660 to 0.723), $88 USD. Among the configurations that score above 0.80, every pair of ranges overlaps. Differences in mean F1, with 95% intervals: Claude Opus 5.5 with Claude Sonnet 5.5 subagents scored 0.4 points higher than Claude Opus 5.5 alone at `medium` (1.1 lower to 1.9 higher), 1.9 lower than alone at `high` (0.6 to 3.2), and 7.1 higher than Claude Sonnet 5.5 alone (4.6 to 9.7). Alone, Claude Opus 5.5 scored 2.3 points higher at `high` than at `medium` (0.7 to 3.9). Claude Fable 5.1 alone scored 1.8 points lower at `low` than at `high` (0.8 to 2.8); at `high` it scored 2.0 points higher than Claude Opus 5.5 alone at `medium` (0.7 to 3.4) and 0.3 lower than Claude Opus 5.5 alone at `high` (1.5 lower to 1.0 higher). With Claude Sonnet 5.5 subagents, Claude Fable 5.1 scored 0.6 points lower than Claude Opus 5.5 (1.9 lower to 0.6 higher). The Claude Opus 5.5 pairing cost 31% less than Claude Opus 5.5 alone at `medium` and 41% less than alone at `high`. Alone, Claude Opus 5.5 cost 15% less at `medium` than at `high`. All seven of November’s runs of Claude Sonnet 5.5 alone scored below October’s three (0.669 to 0.775, compared with 0.783 to 0.788); the cause is not known, and every run counts. Claude Fable 5.1 with Claude Haiku 5.5 subagents cost $85.77 USD per run in October and $89.06 USD in November.

958 * **How priced:** The platform’s list cost of each run’s sessions: tokens with prompt caching, plus session runtime. Five of Claude Opus 5.5’s runs at `high` are priced only from their token counts, at 5-minute cache prices, because each ran partly with the 1-hour cache duration, whose cache writes cost more. The cost of an extra check on tool calls, part of the evaluation’s safety setup, is left out.

959 * **Results:** Mean F1 (range) and mean cost per run. Claude Haiku 5.5 alone: 0.577 (0.550 to 0.606), $25 USD. Claude Sonnet 5.5 alone: 0.760 (0.669 to 0.788), $103 USD. Claude Opus 5.5 alone at `low`: 0.489 (0.437 to 0.551), $38 USD; at `medium`: 0.827 (0.797 to 0.850), $200 USD; at `high`: 0.850 (0.829 to 0.876), $236 USD. Claude Fable 5.1 alone at `low`: 0.829 (0.807 to 0.841), $527 USD; at `medium`: 0.839 (0.817 to 0.856), $603 USD; at `high`: 0.847 (0.829 to 0.861), $716 USD. Claude Fable 5.2 alone at `low`: 0.866 (0.852 to 0.884), $357 USD; at `medium`: 0.877 (0.863 to 0.884), $378 USD; at `high`: 0.909 (0.896 to 0.925), $563 USD. Claude Opus 5.5 with Claude Sonnet 5.5 subagents: 0.831 (0.814 to 0.850), $139 USD; with Claude Haiku 5.5 subagents: 0.677 (0.640 to 0.718), $47 USD. Claude Fable 5.1 with Claude Sonnet 5.5 subagents: 0.825 (0.812 to 0.858), $172 USD; with Claude Haiku 5.5 subagents: 0.687 (0.660 to 0.723), $88 USD. Claude Fable 5.2 with Claude Sonnet 5.5 subagents: 0.833 (0.795 to 0.856), $165 USD; with Claude Haiku 5.5 subagents: 0.716 (0.683 to 0.773), $85 USD. Among the other models’ configurations that score above 0.80, every pair of ranges overlaps. Differences in mean F1, with 95% intervals: Claude Opus 5.5 with Claude Sonnet 5.5 subagents scored 0.4 points higher than Claude Opus 5.5 alone at `medium` (1.1 lower to 1.9 higher), 1.9 lower than alone at `high` (0.6 to 3.2), and 7.1 higher than Claude Sonnet 5.5 alone (4.6 to 9.7). Alone, Claude Opus 5.5 scored 2.3 points higher at `high` than at `medium` (0.7 to 3.9). Claude Fable 5.1 alone scored 1.8 points lower at `low` than at `high` (0.8 to 2.8); at `high` it scored 2.0 points higher than Claude Opus 5.5 alone at `medium` (0.7 to 3.4) and 0.3 lower than Claude Opus 5.5 alone at `high` (1.5 lower to 1.0 higher). With Claude Sonnet 5.5 subagents, Claude Fable 5.1 scored 0.6 points lower than Claude Opus 5.5 (1.9 lower to 0.6 higher). Against Claude Opus 5.5 alone at `medium` and at `high`, Claude Fable 5.2 alone scored higher at each level, differences taken before rounding: at `low`, by 3.9 points (2.4 to 5.3) and 1.6 (0.3 to 2.9), for 1.8 and 1.5 times the cost; at `medium`, by 5.0 (3.6 to 6.3) and 2.7 (1.5 to 3.8), for 1.9 and 1.6 times; at `high`, by 8.2 (6.8 to 9.6) and 5.9 (4.7 to 7.1), for 2.8 and 2.4 times. It scored higher than Claude Fable 5.1 alone at the same level: by 3.6 points at `low` (2.6 to 4.7), 3.8 at `medium` (2.8 to 4.8), and 6.1 at `high` (5.3 to 7.0), for 32%, 37%, and 21% less. It scored 4.3 points lower at `low` than at `high` (3.3 to 5.3), for 37% less. With Claude Sonnet 5.5 subagents, Claude Fable 5.2 scored 0.2 points higher than Claude Opus 5.5 (1.3 lower to 1.7 higher), for 19% more, and scored 3.2 points lower than Claude Fable 5.2 alone at `low` (1.8 to 4.7), 4.3 lower than at `medium` (3.0 to 5.7), and 7.5 lower than at `high` (6.1 to 9.0), for 54%, 56%, and 71% less. One run of Claude Fable 5.2 alone at `high` ended early, without a score, and was replaced. The Claude Opus 5.5 pairing cost 31% less than Claude Opus 5.5 alone at `medium` and 41% less than alone at `high`. Alone, Claude Opus 5.5 cost 15% less at `medium` than at `high`. All seven of November’s runs of Claude Sonnet 5.5 alone scored below October’s three (0.669 to 0.775, compared with 0.783 to 0.788); the cause is not known, and every run counts. Claude Fable 5.1 with Claude Haiku 5.5 subagents cost $85.77 USD per run in October and $89.06 USD in November.

960 * **How priced:** The platform’s list cost of each run’s sessions: tokens with prompt caching, plus session runtime. Five of Claude Opus 5.5’s runs at `high` are priced only from their token counts, at 5-minute cache prices, because each ran partly with the 1-hour cache duration, whose cache writes cost more. Claude Fable 5.2’s own tokens are priced from their counts, at list prices with 5-minute cache writes; in its runs alone, the runtime of its sessions, under 0.2% of a run’s cost, is left out. The cost of an extra check on tool calls, part of the evaluation’s safety setup, is left out.

959961

9609629. **GPQA Diamond:** Rein et al., “GPQA: A Graduate-Level Google-Proof Q\&A Benchmark,” 2023. The 198-question Diamond subset, model-graded against reference answers, advisor tokens metered per request. Claude Haiku 5.5 and Claude Sonnet 5.5, each alone and with a Claude Opus 5.5 advisor, and Claude Opus 5.5 alone ran two runs per configuration on October 7, 2026, all with [server-side fallback](https://platform.claude.com/docs/en/build-with-claude/refusals-and-fallback#server-side-fallback) off, at each model’s default effort, and with refusals counted as wrong. A platform safety check refused one biology question per run on Claude Haiku 5.5 and Claude Sonnet 5.5 alone, one in two runs on Haiku 5.5 with the advisor, none on Sonnet 5.5 with the advisor, and six per run on Opus 5.5, which scored 91% with them counted as wrong and 93.5% on the questions it answered. Claude Haiku 5.5 alone scored 85%, 6 points below Opus 5.5. The advisor configurations added the tool without a prompt asking for consultations, and neither executor called the advisor on any question. The pairings scored 0.5 points (Claude Haiku 5.5) and 1.0 points (Claude Sonnet 5.5) above the executor alone, both within run-to-run noise.

961963

from line 970

968970 * **How priced:** Costs are per attempt. Costs are priced as a customer’s organization is metered: each agent-loop request’s prior prompt as a cache read and its new tokens as a 5-minute cache write, from the runs’ own usage records, and each advisor call, which uses no cache, from its recorded tokens, all at list prices.

969971 * **Claude Opus 5.5 alone:** Claude Opus 5.5 alone ran on all 370 tasks, September 19 to 20, 2026: at its default effort (`medium`) and at `high` with five attempts per task, and at `low` and `xhigh` with one (369 of 370 scored at each, after a setup-check failure). Claude Opus 5.5 alone solved 74.8% of attempts at `low` for $0.35 USD per attempt, 86.6% at `medium` for $0.84 USD, 88.4% at `high` for $1.38 USD, and 91.1% at `xhigh` for $4.11 USD.

970972 * **Claude Opus 5.5 with a Claude Fable 5.1 advisor, runs:** The Claude Opus 5.5 executor at `high` with the released Claude Fable 5.1 as advisor (the August runs used a pre-release snapshot) ran five attempts per task on the same dates; one task failed its setup check, so 1,845 attempts were scored. The 279 attempts in which the advisor was turned away under load were re-run, and attempts whose consults timed out were kept, as in August.

971 * **Claude Fable 5.1 alone:** For comparison, the released Claude Fable 5.1 alone at `high`, its default, scored 85.7% (317 of 370 tasks), one attempt per task on October 7, 2026.

973 * **Claude Fable 5.1 and Claude Fable 5.2 alone:** For comparison, the released Claude Fable 5.1 alone at `high`, its default, scored 85.7% (317 of 370 tasks), one attempt per task on October 7, 2026. Claude Fable 5.2 alone ran on all 370 tasks, one attempt per task at `low`, `medium`, and `high` (369 of 370 scored at `medium`, after a setup-check failure).

972974 * **Consultations:** The August pairing averaged about two advisor consultations per attempt; the Claude Opus 5.5 pairing requested 1.39 and received 1.35.

973975 * **Claude Opus 5.5 with a Claude Fable 5.1 advisor, results:** The Claude Opus 5.5 pairing scored 90.1% at $2.92 USD per attempt: compared with Claude Opus 5.5 at its default, `medium`, that is 3.5 points for about 3.5 times the money. The latency cost is the consultations themselves: about one or two extra frontier-model calls per task, each on the task’s critical path. With the Claude Opus 5.5 executor, the advice saved almost no executor cost: $1.36 USD per attempt compared with $1.38 USD for Claude Opus 5.5 alone at `high`, while the consultations cost $1.55 USD.

974976 * **August pairing:** In August, with a Claude Opus 5 executor and a Claude Fable 5.1 advisor, the executor spent $1.26 USD less per attempt than Claude Opus 5 alone at its default, and the consultations cost $2.47 USD, so that saving paid for about half of the advice. That August pairing was the most accurate configuration measured at the time, at $6.21 USD per attempt.

975977 * **Claude Sonnet 5.5 alone, runs:** Claude Sonnet 5.5 alone ran on all 370 tasks on September 24, 2026, at its default effort (`high`) and at `medium`, one attempt per task, with the same 128,000-token output cap, which cut off none of its attempts. All 370 were scored at `medium`, and 369 of 370 at `high`, after a setup-check failure. Its costs are priced as a customer’s organization is metered.

976978 * **Claude Sonnet 5.5 alone, results:** With no advisor, Claude Sonnet 5.5 scored 77.8% at its default, `high`, for $0.33 USD per attempt, about 9 points below Claude Opus 5.5 at its default for about two fifths of the cost; at `medium`, it scored 62.7% for $0.16 USD. At `medium`, Claude Sonnet 5.5 scored about 15 points lower than at its default, `high`, for about 48% of the cost per attempt.

977 * **Compare models:** In [Compare models](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#compare-models-on-cost-per-task), Claude Opus 5.5 at its default cost under a third as much per attempt as Claude Fable 5.1 at `medium`, which is a single run.

979 * **Compare models:** In [Compare models](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#compare-models-on-cost-per-task), Claude Opus 5.5 at its default cost under two thirds as much per attempt as Claude Fable 5.2 at `medium`, which is a single run. At `low`, Fable 5.2 also scored 88.9% for $1.05 USD per attempt. At `high`, its default on the Messages API, it scored 93.5% for $5.39 USD per attempt.

978980

97998112. **Internal repository-task benchmark (cap measurement):**

980982

981 * **What was measured:** A separate Anthropic-internal set of about 130 repository tasks, run August 20, 2026 (Claude Fable 5.1) and September 19, 2026 (Claude Opus 5.5, at its default effort, `medium`), with a plain API agent loop, one attempt per task.

983 * **What was measured:** A separate Anthropic-internal set of about 130 repository tasks, run August 20, 2026 (Claude Fable 5.1) and September 19, 2026 (Claude Opus 5.5, at its default effort, `medium`), with a plain API agent loop, one attempt per task. The Claude Fable 5.2 runs used the same loop, one attempt per task; in them, as in the Claude Opus 5.5 runs, a turn that reached the cap ended its attempt.

982984 * **How priced:** Cost per solved task is priced as a customer is billed, with a 5-minute prompt cache, for Claude Fable 5.1, and at list prices from the token counts its runs recorded for Claude Opus 5.5.

983 * **Claude Fable 5.1 runs:** The Claude Fable 5.1 runs are 135 tasks per cap at the default effort set explicitly: the 16,384-token figure averages two runs (36.3% on both); the 64,000 and 128,000 figures are single runs (58.5% and 60.0%). Five problems drew a safety refusal in all four runs and count as failures.

985 * **Claude Fable 5.1 and Claude Fable 5.2 runs:** The Claude Fable 5.1 runs are 135 tasks per cap at the default effort set explicitly. The Claude Fable 5.2 runs are 134 tasks per cap at `high`, its default on the Messages API, set explicitly; one task was withdrawn from the set after the Claude Fable 5.1 and Claude Opus 5.5 runs. One further task failed the benchmark’s own check of its reference solution before the model started, in every run, so it is left out and 133 tasks are scored per cap. The 16,384-token figure averages two runs (19.5% and 21.1%); the 64,000 and 128,000 figures are single runs (77.4% and 78.2%). At 64,000, one task failed that check on its first run and was run again; it passed the check and is scored. No Fable 5.2 attempt at 16,384 or 64,000 drew a safety refusal. At 128,000, one attempt drew a safety refusal and one was stopped by a safety monitor in the evaluation environment before it finished; both count as failures. The changes the stopped attempt had made by then passed the tests; leaving it out would give 78.8% at 128,000. The Claude Fable 5.2 and Claude Opus 5.5 costs in the table price each attempt’s recorded token usage at the model’s list prices. The two models’ results were scored by different versions of the benchmark’s grading code.

984986 * **Claude Opus 5.5 runs:** The Claude Opus 5.5 16,384-token figure averages two runs (134 and 135 tasks scored), and its 64,000 and 128,000 figures are single runs (135 tasks each); two attempts in each 16,384-token run and one in each of the 64,000- and 128,000-token runs ended in a safety refusal and count as failures. Claude Opus 5.5 solved 57.6% at 16,384 (58.2% and 57.0%), 75.6% at 64,000, and 68.9% at 128,000. None of its attempts was cut off at 64,000 or 128,000, so the cap does not explain the difference between those two runs.

985 * **Attempts the cap ended:** Only 1 of the 66 Claude Opus 5.5 attempts that the 16,384-token cap ended, and 9 of the 117 Claude Fable 5.1 attempts, still passed.

986 * **Cost per attempt:** Cost per attempt was $3 USD at 16,384 and $4 USD at 64,000 for Claude Opus 5.5, at list prices, and $8 USD and $13 USD for Claude Fable 5.1, as a customer is billed.

987 * **Attempts the cap ended:** Only 1 of the 66 Claude Opus 5.5 attempts that the 16,384-token cap ended, and 9 of the 117 Claude Fable 5.1 attempts, still passed. So did 20 of the 226 Claude Fable 5.2 attempts.

988 * **Cost per attempt:** Cost per attempt was $3 USD at 16,384 and $4 USD at 64,000 for Claude Opus 5.5, at list prices, and $8 USD and $13 USD for Claude Fable 5.1, as a customer is billed. It was $6 USD and $19 USD for Claude Fable 5.2, at list prices.

987989 * **SWE-bench Pro cap figures:** The SWE-bench Pro cap figures are one Claude Fable 5.1 run per cap at the default effort, run August 26, 2026, on a 100-problem subset stratified from reference 3’s 482-problem set, not comparable to its scores; at the default effort the runs at 16,384 and at 64,000 each solved 94 of the 100 problems.

988 * **Turn length:** The per-turn distributions in the section’s dot plot come from the Claude Opus 5.5 and Claude Fable 5.1 runs at 128,000: no Opus 5.5 turn reached the cap (the longest was about 61,000 tokens, and 0.56% of its turns exceeded 16,384), and one Fable 5.1 turn reached 128,000 (0.46% of its turns exceeded 16,384). With the cap at 128,000, 2 of Claude Fable 5.1’s 14,331 turns were longer than 64,000 tokens, and no Claude Opus 5.5 turn was.

990 * **Turn length:** The per-turn distributions in the section’s dot plot come from the Claude Opus 5.5 and Claude Fable 5.1 runs at 128,000: no Opus 5.5 turn reached the cap (the longest was about 61,000 tokens, and 0.56% of its turns exceeded 16,384), and one Fable 5.1 turn reached 128,000 (0.46% of its turns exceeded 16,384). With the cap at 128,000, 2 of Claude Fable 5.1’s 14,331 turns were longer than 64,000 tokens, and no Claude Opus 5.5 turn was. With the cap at 128,000, 303 of Claude Fable 5.2’s 17,750 turns (1.7%) were longer than 16,384 tokens and 7 were longer than 64,000, counting the turns of attempts that were canceled and run again.

989991

99099213. **Chartography:** Surge AI, “Chartography,” 2026.

991993

from line 1050

10481050 * **Compaction trigger on Claude Haiku 5.5:** Measured October 21, 2026, on the Messages API with [compaction at a token threshold](https://platform.claude.com/docs/en/build-with-claude/compaction-threshold), no tools, adaptive thinking, the default effort (`medium`), the 5-minute cache, a cache breakpoint on the system prompt, and [automatic caching](https://platform.claude.com/docs/en/build-with-claude/prompt-caching#automatic-caching). Each session was one conversation of 45 requests: a 5,000-token system prompt, 10,000 tokens of filler text with every user turn, and one-word assistant turns, except that a reply with a compaction block was sent back whole. Four settings ran three sessions each: triggers of 150,000 tokens (the default), 90,000, and 50,000 (the lowest allowed), and no compaction. Costs within a setting agreed to within 4%. With the default trigger, 15 of each session’s 45 replies and all 3 summaries had prompts over 100,000 tokens. With a trigger of 90,000, the largest prompt held 97,269 tokens. Compaction ran at the first request whose whole prompt had reached the trigger, counting the system prompt, cache reads, and cache writes. The summary was written from that prompt plus 44 tokens, which together were up to 7,269 tokens over the trigger. Three more sessions, with a 22,000-token system prompt and a trigger of 50,000, showed that the system prompt counts. The summaries were short (184 to 640 output tokens) because the text was filler. The requests ran in one Anthropic-internal organization on the Claude API.

10491051 * **Compaction trigger on Claude Haiku 5.5, how priced:** From the `usage` fields at list prices, a session cost $0.272 USD with the default trigger, $0.082 USD at 90,000, $0.075 USD at 50,000, and $0.729 USD with no compaction. Per request, from the request after the first compaction through the last compaction, the three triggers cost $0.00605 USD, $0.00181 USD, and $0.00166 USD. With every prompt at the prices for prompts up to 100,000 tokens, the default would cost $0.00206 USD per request, and a trigger of 90,000 would save 12%. A compaction’s summary and reply are each priced by their own prompt’s size.

10501052

105119. **Cache duration measurement on Claude Fable 5.1:** The same 20-issue triage job and harness as reference 16, with `max_tokens` raised to 4,096, run August 23 and August 26, 2026, on the Claude Fable 5.1 launch snapshot at its launch prices ($10 USD input, $12.50 USD 5-minute write, $20 USD 1-hour write, $0.25 USD cache read, $50 USD output per million tokens), three settings per schedule: the 5-minute cache, the 1-hour cache, and the 5-minute cache kept warm by a `max_tokens: 0` request on the unchanged prefix every 4 minutes, timed from the previous request’s start (the August 23 runs sent keep-alive requests with `max_tokens: 1`; in the August 26 cells reported here, every answered keep-alive request refreshed the cache and its response reported no output tokens). Schedules: no pauses, 10% of turns, and every turn at 6 minutes on all 20 issues, and 45-minute pauses on the 5-issue subset; three runs per cell (six for the August 26 keep-alive cell with 45-minute pauses), cost computed from each response’s `usage` fields at list prices, accuracy against the same gold labels (12 to 18 exact labels of 20). Per-session means on August 26 for the 5-minute, 1-hour, and keep-alive settings: no pauses $2.42 USD, $3.09 USD, $2.29 USD; 10% paused $4.50 USD, $2.96 USD, $2.36 USD; every turn $22.89 USD, $3.01 USD, $2.62 USD; the August 23 cells agree within 6%. The 45-minute figures ($1.68 USD, $0.59 USD, and $0.71 USD per 5-issue session) are from later sessions of August 26: that day’s first set of sessions was cut off when its machine was shut down, and the second was affected by a fault in the service and is set aside. The third set gives the 5-minute and 1-hour figures; its three keep-alive sessions sent each keep-alive request once, lost requests to rate-limit refusals, and are not in the mean. The keep-alive figure is the mean of six further sessions that sent any refused keep-alive request again within two seconds. The August 23 runs gave $1.67 USD, $0.58 USD, and $0.70 USD. The break-even between the 5-minute and 1-hour settings is 3.1% of turns, the same measure as reference 16. Claude Fable 5.1’s cache read costs 0.025x the input price while its cache writes keep the standard multipliers, so a keep-alive request that re-reads the prefix is cheap and the 1-hour duration’s write premium is the larger bill, except with 45-minute pauses, where the keep-alive requests cost more than the premium.

105319. **Cache duration measurement on Claude Fable 5.1 and Claude Fable 5.2:** The same 20-issue triage job and harness as reference 16, with `max_tokens` raised to 4,096, run August 23 and August 26, 2026, on the Claude Fable 5.1 launch snapshot at its launch prices ($10 USD input, $12.50 USD 5-minute write, $20 USD 1-hour write, $0.25 USD cache read, $50 USD output per million tokens), three settings per schedule: the 5-minute cache, the 1-hour cache, and the 5-minute cache kept warm by a `max_tokens: 0` request on the unchanged prefix every 4 minutes, timed from the previous request’s start (the August 23 runs sent keep-alive requests with `max_tokens: 1`; in the August 26 cells reported here, every answered keep-alive request refreshed the cache and its response reported no output tokens). Schedules: no pauses, 10% of turns, and every turn at 6 minutes on all 20 issues, and 45-minute pauses on the 5-issue subset; three runs per cell (six for the August 26 keep-alive cell with 45-minute pauses), cost computed from each response’s `usage` fields at list prices, accuracy against the same gold labels (12 to 18 exact labels of 20). Per-session means on August 26 for the 5-minute, 1-hour, and keep-alive settings: no pauses $2.42 USD, $3.09 USD, $2.29 USD; 10% paused $4.50 USD, $2.96 USD, $2.36 USD; every turn $22.89 USD, $3.01 USD, $2.62 USD; the August 23 cells agree within 6%. The 45-minute figures ($1.68 USD, $0.59 USD, and $0.71 USD per 5-issue session) are from later sessions of August 26: that day’s first set of sessions was cut off when its machine was shut down, and the second was affected by a fault in the service and is set aside. The third set gives the 5-minute and 1-hour figures; its three keep-alive sessions sent each keep-alive request once, lost requests to rate-limit refusals, and are not in the mean. The keep-alive figure is the mean of six further sessions that sent any refused keep-alive request again within two seconds. The August 23 runs gave $1.67 USD, $0.58 USD, and $0.70 USD. The break-even between the 5-minute and 1-hour settings is 3.1% of turns, the same measure as reference 16. Claude Fable 5.1’s cache read costs 0.025x the input price while its cache writes keep the standard multipliers, so a keep-alive request that re-reads the prefix is cheap and the 1-hour duration’s write premium is the larger bill, except with 45-minute pauses, where the keep-alive requests cost more than the premium. Claude Fable 5.2 ran the same job at `high`, its default on the Messages API, and at list prices ($8 USD input, $10 USD 5-minute write, $16 USD 1-hour write, $0.20 USD cache read, $40 USD output per million tokens). Like the Claude Opus 5.5 cells in reference 16, it used a port of the harness that sends the same request bodies, with `max_tokens` raised to 4,096. Schedules: no pauses, 5% and 10% of turns, and every turn at 6 minutes on all 20 issues, and 20-minute and 45-minute pauses on the 5-issue subset, three runs per cell. The 45-minute cells were measured twice; their figures pool both sets of three sessions and, as with Claude Fable 5.1, leave out the keep-alive sessions that lost keep-alive requests (here, to timeouts; one in each set), so the 45-minute keep-alive figure is the mean of four sessions and the 5-minute and 1-hour figures the mean of six. Throughout, the keep-alive requests were sent with `max_tokens: 0`. The test client waits 30 seconds for a keep-alive request and does not retry. Across the schedules listed above, it stopped waiting for 3 of 886 keep-alive requests, all in the 45-minute cells, and each time the cache expired and the next keep-alive request wrote the prefix again. Per-session means for the 5-minute, 1-hour, and keep-alive settings: no pauses $2.18 USD, $2.66 USD, $2.23 USD; 10% paused $4.16 USD, $2.82 USD, $2.29 USD; every turn at 6 minutes $20.36 USD, $2.61 USD, $2.50 USD; and $1.43 USD, $0.53 USD, and $0.65 USD per 5-issue session with 45-minute pauses. The Claude Fable 5.2 savings in [Pick the cache duration](https://platform.claude.com/docs/en/about-claude/models/optimizing-for-cost-and-intelligence#pick-the-cache-duration) are taken before rounding. With the two sessions that lost keep-alive requests included, the 45-minute keep-alive mean was $0.73 USD, $0.20 USD more than the 1-hour setting.

10521054

1053105520. **Terminal-Bench 3:** the public terminal-agent benchmark’s 74 tasks, run on [Claude Managed Agents](https://platform.claude.com/docs/en/managed-agents/overview) with two custom tools, a shell and a file editor that the evaluation harness runs in each task’s own container, in place of the platform’s built-in tools, and otherwise at the platform’s default settings for external accounts, two runs per model at `high` effort, August 27 to 28, 2026. These runs used Terminal-Bench version 3.0, and their scores are not comparable with the public Terminal-Bench leaderboard or with the Terminal-Bench 4.0 results in the Claude Opus 5.5 system card, which come from runs in Claude Code at `max` effort. Each task’s time limits are 2.5 times the benchmark’s own, which gives the agent between 75 minutes and 20 hours per task (5 hours for the median task), and each task gets three times the memory it specifies, from 6 GiB to 96 GiB, with extra memory for the 12 tasks that run helper services. The agent had no general internet access: its containers could reach an internal package mirror, a short list of download sites including GitHub and the Python Package Index, and a few sites specific to some tasks, and eight of the tasks had no network access at all. Scores are raw pass rates over the 148 attempts per model; single runs swing by 5 to 11 points. Costs are what a customer would be billed at list prices, re-priced request by request from the runs’ usage records with the 5-minute cache lifetime. Claude Opus 4.7, Claude Opus 4.8, and Claude Opus 5 each spend $8 USD to $15 USD per task. Claude Opus 4.7 ended 11 of its 148 attempts at its output cap.

10541056

This page’s history 36 edits

11 Nov 2026 +32 −60
9 Nov 2026 +85 −83 · this one
5 Nov 2026 +37 −31
4 Nov 2026 +5 −3
30 Oct 2026 +101 −101
29 Oct 2026 +32 −35

The page’s own history →

Where it came from

The captureapi
Upstream pageplatform.claude.com

Around it

Every changethe ledger
That day9 Nov 2026