feat(providers): add Claude Sonnet 5.5 and make it the default model

[waleedlatif1]

Summary

• Add claude-sonnet-5-5: $2/$10 per MTok, $0.20 cache read, 1M context, 128K max output, adaptive thinking (low–max, default high), native structured outputs, 512-token cache minimum
• forcedToolUse: false: Sonnet 5.5 rejects tool_choice tool/any with a 400, so a forced tool is sent as auto (and the Force option is hidden in tool input), same as Opus 5.5 / Fable 5.1
• Map the none thinking level to thinking: { type: "between_tools" } via a new capabilities.thinking.noneMode field. Sonnet 5.5 rejects disabled and is adaptive by default, so previously none silently ran full adaptive thinking at high. between_tools carries no effort/display (both 400). Other models keep sending no thinking config for none
• Entry sits before claude-sonnet-5 so the prefix-matching catalog lookups (startsWith(id + '-')) resolve it to its own capabilities
• Make Sonnet 5.5 the recommended model and the default for the Agent, Router, and Evaluator blocks, the combobox model fallback, and the Anthropic provider; mark claude-sonnet-5 legacy (Anthropic now lists it as legacy). Router/Evaluator use responseFormat (native structured outputs), not forced tools
• Update docs defaults and regenerate the agent streaming table

Type of Change

• New feature

Testing

• Live against the Anthropic API: Models API confirms 1M input / 128K output / 5 effort levels / structured outputs / adaptive only; raw probes confirm disabled, enabled, non-default temperature, forced/any tool_choice, and between_tools + xhigh/display all 400, while every shape Sim sends returns 200; a 683-token prompt caches
• Live end-to-end through executeAnthropicProviderRequest with the real SDK: none + tools (streaming and non-streaming tool loops, between_tools on every turn), high + tools streaming with agent events (adaptive summarized, thinking-block replay accepted), forced tool downgraded to auto, router-style responseFormat with temperature set, xhigh/max, streaming without tools, and prompt-cache write then read
• providers/anthropic/core.test.ts: forced-tool and none → between_tools tests, both confirmed red with the catalog fields reverted
• vitest run providers executor blocks lib/model-router combobox (4,055 tests), bun run type-check, bun run lint, bun run check:audits (51 audits), block registry check, docs-manifest:check
• Pricing cross-checked against Anthropic pricing and OpenRouter

Checklist

• Code follows project style guidelines
• Self-reviewed my changes
• Tests added/updated and passing (new tests pass the test-audit authoring gate)
• No new warnings introduced
• I confirm that I have read and agree to the terms outlined in the Contributor License Agreement (CLA)

[vercel]

The latest updates on your projects. Learn more about Vercel for GitHub.

1 Skipped Deployment

Project	Deployment	Actions	Updated
docs	Skipped		Sep 28, 2026 7:53pm UTC

[cubic-dev-ai]

No issues found across 15 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Re-trigger cubic}

[greptile-apps]

Confidence Score: 5/5

[Medium risk] Changes default AI model across the application.

The PR appears safe to merge based on the reviewed changes and resolved prior threads.

Summary

The PR adds Claude Sonnet 5.5 to the model catalog, makes it the default for Agent, Router, and Evaluator blocks, and maps its none thinking selection to between_tools. The latest revision replaces mock-call assertions with recorded request payloads and retains multi-turn coverage. No new actionable issue was identified.

Diagram

%%{init: {'theme': 'neutral'}}%%
flowchart LR
  A[Model selection] --> B[Sonnet 5.5 capabilities]
  B --> C{Thinking level}
  C -->|none| D[between_tools]
  C -->|other supported level| E[adaptive thinking]
  D --> F[Anthropic request]
  E --> F

Loading

_{Reviews (3) · Last reviewed commit: "test(providers): assert between_tools on..."}

[waleedlatif1]

@greptile

[waleedlatif1]

@cubic-dev-ai review this PR

[cubic-dev-ai]

@cubic-dev-ai review this PR

@waleedlatif1 I have started the AI code review. It will take a few minutes to complete.

[cubic-dev-ai]

No issues found across 15 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Re-trigger cubic}

[waleedlatif1]

@greptile

[waleedlatif1]

@cubic-dev-ai review this PR

[cubic-dev-ai]

@cubic-dev-ai review this PR

@waleedlatif1 I have started the AI code review. It will take a few minutes to complete.

[cubic-dev-ai]

No issues found across 15 files

Confidence score: 5/5

• Automated review surfaced no issues in the provided summaries.
• No files require special attention.

_{Re-trigger cubic}