[Feat] Add Claude Opus 5 support across all providers

[zoomote]

Opened on behalf of Naved Merchant. Follow up by mentioning @zoomote, in the web UI, or in Discord.

Related GitHub Issue

No dedicated issue exists for Claude Opus 5 yet. This PR mirrors the approved Claude Sonnet 5 work: #777 / #778.

Description

Adds full Claude Opus 5 support across all provider paths, mirroring how Claude Sonnet 5 was added in #778, with pricing and parameters researched from Anthropic's published docs.

Model definitions (packages/types — anthropic, bedrock, vertex):

• claude-opus-5 (Anthropic/Vertex) and anthropic.claude-opus-5 (Bedrock)
• 1M native context window (no beta header required), 128k max output
• Adaptive-thinking binary toggle (supportsReasoningBudget/supportsReasoningBinary); supportsTemperature: false since sampling params return a 400
• Pricing per Anthropic docs: $5 input / $25 output / $6.25 cache writes / $0.50 cache reads per million tokens
• Bedrock: added to BEDROCK_GLOBAL_INFERENCE_MODEL_IDS

Provider handling:

• Added opus-5 to the Bedrock adaptive-thinking guard (safe against false positives: claude-opus-4-5-* IDs do not contain the literal substring opus-5)
• Added anthropic/claude-opus-5 to the OpenRouter and Vercel AI Gateway prompt-caching, reasoning-budget, and vision/tools allow-lists (confirmed available on OpenRouter)
• Applied Opus 5 overrides (maxTokens, reasoning budget/binary, temperature) in the OpenRouter, Requesty, and Vercel AI Gateway fetchers
• Added claude-opus-5 to both Anthropic handler switch statements (cache breakpoints and prompt-caching beta header)

Test Procedure

• Added Opus 5 test coverage mirroring the Sonnet 5 tests in every touched suite: Anthropic, Anthropic Vertex, Bedrock (model info, global inference prefix, adaptive-thinking request shape, guard predicate), Requesty, Vercel AI Gateway, all three fetcher specs, and the shared API max-output-tokens spec.
• pnpm test passes: 6,948 tests passed, 38 skipped across 424 test files.
• Pre-push lint and check-types hooks pass across all 14 packages.

Pre-Submission Checklist

• Issue Linked: No dedicated issue exists; linked to the analogous approved Sonnet 5 work ([BUG] Implement Claude Sonnet 5 support in Zoo Code #777 / feat: implement Claude Sonnet 5 support in Zoo Code #778) above.
• Scope: My changes are focused on the linked issue (one major feature/fix per PR).
• Self-Review: I have performed a thorough self-review of my code.
• Testing: New and/or updated tests have been added to cover my changes (if applicable).
• Documentation Impact: I have considered if my changes require documentation updates (see "Documentation Updates" section below).
• Contribution Guidelines: I have read and agree to the Contributor Guidelines.

Screenshots / Videos

Not applicable — provider/model-registry change with no UI surface (model becomes selectable in the existing provider pickers through the standard model registries).

Documentation Updates

• No documentation updates are required. (The docs repo covers providers generically and does not maintain per-model lists; the Sonnet 5 change shipped without docs updates as well.)

Additional Notes

Pricing researched from Anthropic's models overview and pricing pages: Opus 5 is $5/$25 per MTok with $6.25/$0.50 cache rates (no introductory pricing, unlike Sonnet 5), adaptive thinking only, no extended-thinking budget_tokens, and no sampling parameters.

Get in Touch

@navedmerchant on GitHub; Naved Merchant on Discord.

[zoomote]

No code issues found. See task

Reviewed the Opus 5 model definitions (Anthropic/Bedrock/Vertex), the Bedrock adaptive-thinking guard, the Anthropic handler switch cases, and the OpenRouter/Requesty/Vercel AI Gateway fetcher overrides against the existing Sonnet 5 / Fable 5 patterns — every location that references claude-sonnet-5 or claude-fable-5 now has a matching claude-opus-5 entry, and the opus-5 substring guard has no false positives among existing Bedrock model IDs. Verified locally: the 9 touched test suites pass (288 tests).

_{Reviewed 4903c40}

[codecov]

Codecov Report

✅ All modified and coverable lines are covered by tests.

📢 Thoughts on this report? Let us know!

[edelauna]

Nice and comprehensive!