Fix model routing and thinking modes

[Aayam Bansal (aayambansal)]

Summary

• deduplicate provider-equivalent model rows while keeping Codex subscription routes separate
• fold OpenRouter fast/pro siblings into service modes and preserve legacy selections
• expose Standard plus provider-native reasoning efforts, independently from Fast/Pro service modes
• send adaptive thinking and effort for current Claude models, while retaining token-budget thinking for older Claude models
• expand browser coverage for independent reasoning and service-mode selection

Root cause

The workspace treated provider aliases and OpenRouter service variants as distinct catalog entries, and the thinking selector had no explicit Standard state. Current Anthropic effort values were also sent without enabling adaptive thinking, so direct Claude requests did not match Anthropic's current request contract.

Validation

• bun run check (1271 pass, 1 skip, 0 fail)
• provider transform tests (121 pass, 0 fail)
• model catalog/tier/variant tests (13 pass, 0 fail)
• Playwright thinking + service-mode flows (2 pass)
• frontend production build
• forced frozen install verifies the pinned Anthropic patch applies
• Prettier check and git diff --check

[vercel]

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

Project	Deployment	Actions	Updated (UTC)
openscience	Ready	Preview, Comment	Jul 27, 2026 3:07pm