Source Intelligence
Follow Discord

DisclaimerUnofficial, and not affiliated with Anthropic. Nearly all of this is read straight out of what ships: npm bundles, captured prompts, published docs. Anthropic's own notes go in verbatim, marked as theirs. The rest is my reading, and every entry carries the strings behind it. If one looks wrong, vote it down and say why.

← All releases Home ←olderv2.1.196 v2.1.198newer→

Claude Code v2.1.197

6 entries read diff v2.1.196 → v2.1.197 Markdown

This release introduces Claude Sonnet 5 (claude-sonnet-5) as the new default Sonnet model for Claude Code, replacing Claude Sonnet 4.6. The model picker gains a promotional pricing display that shows the launch promo price alongside the struck-through list price, and a new "Sonnet 5 (1M context)" option appears alongside the renamed "Sonnet 4.6 (1M context)" entry. Alias-to-canonical-ID resolution was added so that settings referencing saved model IDs survive the transition.

Find
Pick an entry · j / k steps through
3 entries

New Featuresopen

Claude Sonnet 5 Is Now the Default Model#

What

Claude Sonnet 5 (claude-sonnet-5) becomes the default Sonnet tier model, replacing Claude Sonnet 4.6. The sonnet alias now resolves to claude-sonnet-5.

Details
  • Model ID: claude-sonnet-5
  • 1M context window, 64K default output tokens, 128K max output tokens
  • Full adaptive thinking support (max effort level enabled)
  • Available on all providers: first-party Anthropic API, Bedrock (us.anthropic.claude-sonnet-5), Vertex (claude-sonnet-5), Foundry, Gateway, and Mantle (anthropic.claude-sonnet-5)
  • Sonnet 4.6 is retained in the model picker as "Sonnet 4.6 · Previous Sonnet version"
  • Hook examples in settings/CLAUDE.md now reference claude-sonnet-5
Evidence

SONNET_ID constant updated (search for "claude-sonnet-5" in model constants; SONNET_NAME: "Claude Sonnet 5")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Related

Other releases about the same thing. Found by shared names or similar wording; neither means one caused the other.

Promotional Launch Pricing in Model Picker#

What

When Sonnet 5 is on a launch promotion, the model picker displays the promo price with the regular list price struck through next to it.

Details
  • During the promo period, the Sonnet 5 entry shows: ~~$3/$15~~ $2/$10 per Mtok · promo through [date]
  • The promo expiry date is read from the cedar_basin key in the remote server config (AI()?.cedar_basin). Dlo() parses the value and returns it only if Date.now() < parsedMs (live gate); once the date has passed Dlo() returns undefined and no promo UI is rendered
  • Rlo() formats the raw date string using toLocaleDateString("en-US", { month: "short", day: "numeric", timeZone: "UTC" }), producing e.g. "Jun 29"
  • The description field carries the promo price ($2/$10 per Mtok · promo through …) so plain-text consumers read it unambiguously; rich pickers (the CLI model-selector component) additionally prepend chalk.dim.strikethrough(promoListPrice) before the first $X/$Y token in description, producing the struck-through display
  • When no promo is active the description falls back to C1(jte) — the standard pricing description string
  • The "Sonnet 5 (1M context)" entry (Nlo()) follows the same pattern; when running in a standalone/operator context (Ao() returns true) all pricing is omitted and only "Sonnet 5 for long sessions" is shown
  • Third-party provider users (Bedrock, Vertex, etc.) do not see promo pricing: gBn() applies the promo display block only when ud() returns false (first-party API key users)
  • This is a new UI field: promoListPrice added to the model row schema specifically to support struck-through rendering in rich pickers; description always carries only the promo price so plain-text consumers are unaffected
Evidence

Promo gate function Dlo() reading AI()?.cedar_basin; formatter Rlo() (search for "$2/$10 per Mtok" and "promoListPrice"); rich-picker rendering (search for "dim.strikethrough"); schema documentation (search for "@internal List price (e.g. \$3/$15\)")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Sonnet 5 Added to the 1M Context Picker#

What

A "Sonnet 5 (1M context)" option now appears in the extended-context model picker, and the previous "Sonnet (1M context)" entry is renamed to "Sonnet 4.6 (1M context)".

Details
  • New entry: value: "sonnet[1m]" → "Sonnet 5 (1M context)" — "Sonnet 5 with 1M context window - for long sessions with large codebases"
  • Renamed entry: value: "claude-sonnet-4-6[1m]" → "Sonnet 4.6 (1M context)"
  • Sonnet 5 has native_1m support on all 3rd-party providers (Bedrock, Vertex, Foundry, Gateway); availability is checked per-provider against a new native_1m_3p capability flag
  • The "not available for your account" error message updated from "Sonnet 4.6 with 1M context is not available…" to "Sonnet with 1M context is not available…"
Evidence

New picker option (search for "Sonnet 5 (1M context)"); renamed entry (search for "Sonnet 4.6 (1M context)"); availability check function (search for "native_1m_3p")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Related

Other releases about the same thing. Found by shared names or similar wording; neither means one caused the other.

3 entries

Improvementsopen

Alias-to-Canonical-ID Resolution for Persisted Model Preferences#

What

Model alias matching now uses a canonical wire model ID to let hosts correctly match a stored explicit model ID (e.g., a saved claude-sonnet-4-6) against the alias row that covers it, surviving model transitions without breaking saved settings.

Details
  • New resolvedModel field (previously documented as canonicalId) in the model row schema: "Canonical wire model id this row's value resolves to (e.g. 'sonnet' → 'claude-sonnet-5'). Lets hosts match a persisted explicit id against the alias row that covers it." — populated by Y4t() via Bo(value), which resolves an alias to its concrete wire model ID
  • New cda(options, persistedId) function: if persistedId already exists verbatim in the option list it is returned unchanged; otherwise a z4t() scan finds a covering alias by checking in order — exact value match, fable-family membership, same model family AND (same 1M-context flag OR model is natively 1M-capable)
  • When cda() finds a match it returns the alias option's value (e.g., "sonnet"), so the picker highlights the correct alias row for the persisted model; if no alias covers the persisted ID the call-site expression cda(options, initial) ?? initial keeps the original persisted value as the initial selection rather than discarding it
Evidence

Schema field resolvedModel and description (search for "Canonical wire model id this row's"); cda(e, t) function definition; z4t() matcher (search for "Lets hosts match a persisted explicit id")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Explicit Error for Missing Bedrock Fallback ID#

What

The function that resolves default 3rd-party model IDs now throws a descriptive error if a model tier is configured with bedrock: null, instead of silently passing null downstream.

Details
  • Previous behavior: null was passed as the Bedrock fallback ID, which could produce cryptic downstream failures
  • New behavior: immediately throws "A DEFAULT_3P_*_KEY points at a model config with bedrock: null — Bedrock setup has no fallback id for that tier"
  • This surfaces misconfiguration at the resolution point rather than at API call time
Evidence

Error message (search for "A DEFAULT_3P_*_KEY points at a model config with bedrock: null")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Model Documentation Updated in API Reference Context#

What

The built-in API reference context used by the claude-api skill has been updated to reflect the Claude 5 family.

Details
  • Model Migration Guide updated with Sonnet 5 migration guidance including: unchanged features (1M context window, 128K max output, prompt caching, batch, Files API, PDF, vision, tools), plus Sonnet 5-specific notes
  • Model Catalog updated: Sonnet 5 listed as active; Sonnet 4.6 reclassified as "Previous-generation Sonnet" with max output corrected to 128K (was shown as 64K in previous version's docs)
  • Hook examples in the agent/prompt hook documentation now default to claude-sonnet-5
  • The model migration table now includes claude-sonnet-5 in the source column
Evidence

Model catalog entry (search for "Claude Sonnet 5" in model catalog strings); migration table (search for "claude-sonnet-5" / "claude-opus-4-8")

Strings lifted out of the shipped bundle, so the claim above can be checked against them.

Verbatim
Official · Anthropic

Anthropic’s official release notes

Published verbatim by Anthropic for v2.1.197. Text is unmodified from the upstream changelog. Everything else on this page came out of the bundle instead, which is why the two lists don't match.

  • Introducing Claude Sonnet 5: now the default model in Claude Code, with a native 1M-token context window and promotional pricing of $2/$10 per Mtok through August 31. Update to version 2.1.197 for access. https://www.anthropic.com/news/claude-sonnet-5
Read it on GitHub →
System prompt

3 added and 3 removed, of 223 lines, about 124 words, in the prompt 17 of 27 arms receive. 2 other prompts also changed. 1 of 27 tool descriptions changed.

Claude Code, interactive mode

Read the prompts for this release →