Describe the bug
Claude Sonnet 5 is not yet fully implemented in Zoo Code. The model needs to be added across all provider paths (Anthropic direct, Amazon Bedrock, Google Vertex, OpenRouter, Requesty, and Vercel AI Gateway) with correct model definitions, adaptive-thinking handling, fetcher overrides, and accurate pricing.
Per Anthropic's published documentation, Claude Sonnet 5:
- Has a native 1M token context window (no beta header required)
- Uses the newer adaptive-thinking contract (
thinking.type: "adaptive" + output_config.effort), removing sampling parameters (temperature/top_p/top_k) and budget_tokens-based thinking — sending these returns a 400
- Has introductory pricing of $2 / $10 per million input/output tokens in effect through August 31, 2026 (standard $3 / $15 rates do not take effect until September 1, 2026)
- Supports prompt caching with cache writes at $2.50/MTok (1.25× base) and cache reads at $0.20/MTok (0.1× base) during the introductory period
Reference: https://platform.claude.com/docs/en/about-claude/pricing
To Reproduce
Steps to reproduce the behavior:
- Attempt to select
claude-sonnet-5 on any provider (Anthropic, Bedrock, Vertex, OpenRouter, Requesty, or Vercel AI Gateway).
- Observe that the model is missing from model lists, lacks adaptive-thinking handling, and/or reports incorrect token costs (standard $3/$15 instead of introductory $2/$10).
- Sending requests with temperature set returns a 400 from Anthropic because the adaptive-thinking guard does not yet cover Sonnet 5.
Expected behavior
Zoo Code should fully support Claude Sonnet 5 across all providers:
- Model definitions with 1M context window, adaptive-thinking binary toggle, and
supportsTemperature: false
- Correct introductory pricing ($2 input / $10 output / $2.50 cache writes / $0.20 cache reads per MTok) through August 31, 2026
- Inclusion in OpenRouter and Vercel AI Gateway model allow-lists
- Adaptive-thinking guard coverage in the Bedrock provider
- Fetcher overrides (maxTokens, reasoning budget, temperature) for OpenRouter, Requesty, and Vercel AI Gateway
- Test coverage across provider and fetcher test suites
Screenshots
N/A
Video
N/A
What version of zoo are you running
Latest (main)
Additional context
Implementation spans:
packages/types/src/providers/ — anthropic.ts, bedrock.ts, vertex.ts (model definitions + pricing), openrouter.ts, vercel-ai-gateway.ts (model allow-lists)
src/api/providers/ — anthropic.ts, bedrock.ts (adaptive-thinking handling), fetchers/openrouter.ts, fetchers/requesty.ts, fetchers/vercel-ai-gateway.ts (model overrides)
- Test suites across
src/api/providers/__tests__/ and src/api/providers/fetchers/__tests__/
Describe the bug
Claude Sonnet 5 is not yet fully implemented in Zoo Code. The model needs to be added across all provider paths (Anthropic direct, Amazon Bedrock, Google Vertex, OpenRouter, Requesty, and Vercel AI Gateway) with correct model definitions, adaptive-thinking handling, fetcher overrides, and accurate pricing.
Per Anthropic's published documentation, Claude Sonnet 5:
thinking.type: "adaptive"+output_config.effort), removing sampling parameters (temperature/top_p/top_k) and budget_tokens-based thinking — sending these returns a 400Reference: https://platform.claude.com/docs/en/about-claude/pricing
To Reproduce
Steps to reproduce the behavior:
claude-sonnet-5on any provider (Anthropic, Bedrock, Vertex, OpenRouter, Requesty, or Vercel AI Gateway).Expected behavior
Zoo Code should fully support Claude Sonnet 5 across all providers:
supportsTemperature: falseScreenshots
N/A
Video
N/A
What version of zoo are you running
Latest (main)
Additional context
Implementation spans:
packages/types/src/providers/—anthropic.ts,bedrock.ts,vertex.ts(model definitions + pricing),openrouter.ts,vercel-ai-gateway.ts(model allow-lists)src/api/providers/—anthropic.ts,bedrock.ts(adaptive-thinking handling),fetchers/openrouter.ts,fetchers/requesty.ts,fetchers/vercel-ai-gateway.ts(model overrides)src/api/providers/__tests__/andsrc/api/providers/fetchers/__tests__/