Sweep 19 Sep 2026 · 02:36Z Build v2.1.278 500 read Stable v2.1.267 Latest v2.1.278 Next v2.1.278 Feeds RSS JSON llms.txt Unofficial
Claude Code v2.1.276 ·

Model-advisor standby model selection now double-checks the base URL is Anthropic's own

Model-advisor's standby model pick now also requires the account to actually be talking to Anthropic's own API

TierUnder the hoodhow much it should matter to you
Useful2my rating, 1 to 5
Signal2worth watching, 1 to 5
AreaElsewherewhat it touches
KindBug Fixesin v2.1.276,
Under the hood No documentation found

Model-advisor's standby model pick now also requires the account to actually be talking to Anthropic's own API

What

Claude Code has an internal "model-advisor" feature that can automatically switch you to a standby (fallback) model. Previously, this switch only checked that the advisor tool was enabled and that it had ranked a standby model. Now it also checks that the connection is genuinely going to Anthropic's own servers: the account type must be firstParty, and either _CLAUDE_CODE_ASSUME_FIRST_PARTY_BASE_URL is set, or the effective ANTHROPIC_BASE_URL actually resolves to api.anthropic.com.

In practice, this means:

  • If you're signed in with a first-party Anthropic account but have ANTHROPIC_BASE_URL pointed at a custom proxy or gateway, the advisor will no longer silently substitute a standby model.
  • The standby-model switch now only kicks in when requests are verifiably reaching Anthropic's own API endpoint.
Why

This closes a gap where a proxied or gateway-routed session could still get its model silently swapped out by the advisor, even though it wasn't actually talking to Anthropic's servers. Anyone routing Claude Code through a custom ANTHROPIC_BASE_URL should expect the standby-model feature to stay off rather than kick in unexpectedly.

Read from
Names in the bundle_CLAUDE_CODE_ASSUME_FIRST_PARTY_BASE_URL
Feature flag
tengu_sage_compass2 Not enough to say

Nothing here resolved what this flag was doing on this version, so nothing here should be read as on or off.

This account: no value returned · anonymous baseline: no value returned · compiled default in v2.1.276: not a boolean we can read

These values were read against a different version of Claude Code, so treat them as the nearest reading available instead of one taken on this release.

Read once, for one account on one subscription tier, against v2.1.276. It isn't a statement about your account. What a flag value here can and cannot tell you

What the documentation says
api.anthropic.com Configuration reference modified, high confidence Before Claude works with a public site, the app checks its address (query removed) against Anthropic's site safety list on `releases.claude.com`, signed out, with no account, organization, or device identifier (Claude admin console install… see the edit
api.anthropic.com Telemetry and egress modified, medium confidence With `disableEssentialTelemetry`, `disableNonessentialTelemetry`, `disableNonessentialServices`, and `disableAutoUpdates` all set to `true`, and [`modelCatalogEnabled`](/docs/third-party/claude-desktop/configuration#modelcatalogenabled) se… see the edit
How sure we are
Something disagreesSomething we can check disagrees with this entry, or the writer said they could not settle it.
The writer flagged doubtWhether the related `tengu_sage_compass2` gate is enabled for any account is unknown; no reading of it has been taken under this release.
Anthropic's documentation agreesapi.anthropic.com on Configuration reference
Anthropic's release notes agreeFixed every request failing with 400 … Input tag 'advisor_20260301' when ANTHROPIC_BASE_URL points at a proxy or gateway (2.1.275…

See this entry in the whole of v2.1.276 →