Skip to content

fix(anthropic): move (latest) tag from 4.5 to 5 for Opus and Sonnet - #6433

Closed
bharadwaj-pendyala wants to merge 1 commit into
anomalyco:devfrom
bharadwaj-pendyala:fix/stale-latest-tags
Closed

bharadwaj-pendyala wants to merge 1 commit into
anomalyco:devfrom
bharadwaj-pendyala:fix/stale-latest-tags

Conversation

@bharadwaj-pendyala

Copy link
Copy Markdown

Summary

  • claude-opus-4-5 / claude-sonnet-4-5 showed as (latest) in opencode /models picker while newer Opus 5 / Sonnet 5 exist.
  • Moves (latest) display name to the current models per https://docs.anthropic.com/en/docs/about-claude/models (Opus 5 2026-07-24, Sonnet 5 2026-06-30).
  • Haiku 4.5 keeps (latest) — no Haiku 5 exists, it is still the current Haiku.

Changes

  • models/anthropic/claude-opus-4-5.toml: Claude Opus 4.5 (latest) -> Claude Opus 4.5
  • models/anthropic/claude-opus-5.toml: Claude Opus 5 -> Claude Opus 5 (latest)
  • models/anthropic/claude-sonnet-4-5.toml: Claude Sonnet 4.5 (latest) -> Claude Sonnet 4.5
  • models/anthropic/claude-sonnet-5.toml: Claude Sonnet 5 -> Claude Sonnet 5 (latest)

Verification

  • TOML parses cleanly; full bun ./packages/core/script/validate.ts left to CI (no bun locally).
  • Cross-checked current lineup vs Anthropic docs: Fable 5.1 / Opus 5 / Sonnet 5 / Haiku 4.5.

Opus 4.5 and Sonnet 4.5 still carry the (latest) display name while
Opus 5 (2026-07-24) and Sonnet 5 (2026-06-30) are the current models
per https://docs.anthropic.com/en/docs/about-claude/models.

Haiku 4.5 keeps (latest) — no Haiku 5 exists yet.
@github-actions

github-actions Bot commented Sep 6, 2026

Copy link
Copy Markdown
Contributor

Action items

  • [high] [violation] models/anthropic/claude-opus-4-5.toml:1 - Check: (latest) marks undated Anthropic alias IDs, not “newest generation only.” Why: Stripping (latest) from claude-opus-4-5 / claude-sonnet-4-5 conflicts with catalog convention (older undated aliases still use it: claude-opus-4-0, claude-opus-4-1, claude-sonnet-4-0, claude-haiku-4-5) and with Anthropic sync, which always appends (latest) when canonical_id is set (packages/core/src/sync/providers/anthropic.ts, test “labels Anthropic aliases as latest”). Next sync would reintroduce provider-side name = "… (latest)" overrides and undo the lab-only rename. Action: Keep (latest) on the 4.5 undated alias lab entries; only add (latest) to current undated aliases (claude-opus-5, claude-sonnet-5) if those IDs are still floating aliases.
  • [high] [violation] providers/anthropic/models/claude-sonnet-5.toml:1 - Check: First-party Anthropic provider must reflect the intended display name (inherit lab name or override it). Why: This file is a full inline definition with hardcoded name = "Claude Sonnet 5" and no base_model, so the lab rename to Claude Sonnet 5 (latest) never reaches the Anthropic host entry users pick. claude-opus-5 already inherits via base_model; Sonnet 5 does not. Action: Point this file at base_model = "anthropic/claude-sonnet-5" (override-only) or set name = "Claude Sonnet 5 (latest)" to match the lab alias label.
  • [medium] [possible mistake] providers/requesty/models/claude-opus-4-5@eu.toml:2 - Check: Provider display names that bake in (latest) stay consistent with lab alias naming. Why: These Requesty EU overrides still say Claude Opus/Sonnet 4.5 (latest) (EU) while the PR removes that tag from the shared lab metadata; if the PR’s product intent is “only gen 5 is latest,” these stay wrong, and gen-5 EU rows (claude-opus-5@eu, claude-sonnet-5@eu) still omit (latest). Action: Either leave 4.5 lab (latest) in place (preferred with sync/convention), or update the Requesty EU 4.5/5 display names in the same change so they match the chosen rule.

@bharadwaj-pendyala

Copy link
Copy Markdown
Author

Superseded by #6434, which includes the same strips plus the root fix in sync logic so the tag can't rot again.

Sign up for free to join this conversation on GitHub. Already have an account? Sign in to comment

Labels

None yet

Projects

None yet

Development

Successfully merging this pull request may close these issues.

1 participant