Map Anthropic reasoning effort by provider

[xeophon]

Summary

• Map sampling_args.reasoning_effort to extra_body.verbosity for Anthropic models on OpenRouter/Prime OpenAI-compatible routes.
• Map reasoning_effort to native Anthropic output_config.effort and enable adaptive thinking for Claude 4.6/4.7 when unset.
• Add focused client tests for Claude 4.6/4.7 OpenRouter-style and native Anthropic mappings.

Verification

• uv run ruff check --fix verifiers/clients/openai_chat_completions_client.py verifiers/clients/anthropic_messages_client.py tests/test_client_multimodal_types.py
• uv run pytest tests/test_client_multimodal_types.py
• Pre-push hooks: ruff check, ruff format, ty (ci parity)
• Live Prime GPQA 1x1 checks on anthropic/claude-opus-4.7:
  • reasoning_effort="high": https://app.primeintellect.ai/dashboard/evaluations/rbej739yowcedihefu7sqykk
  • reasoning_effort="xhigh": https://app.primeintellect.ai/dashboard/evaluations/gl5n8e3cv0fcesfhjwgm1bcz
  • reasoning_effort="max": https://app.primeintellect.ai/dashboard/evaluations/ieaij5d5dpnjtcmlycby2wjg

---

Note

Medium Risk
Adjusts request-parameter normalization for Anthropic/Claude models across OpenAI-compatible and native Anthropic clients, which can change generation behavior and provider compatibility for affected routes.

Overview
Adds provider-aware handling of sampling_args.reasoning_effort.

For OpenAI-compatible Anthropic/Claude models on OpenRouter/Prime (openrouter.ai/pinference.ai), reasoning_effort is now translated into extra_body.verbosity and ensures extra_body.reasoning.enabled is set, instead of sending reasoning_effort directly.

For the native Anthropic Messages client, reasoning_effort is mapped to output_config.effort, and Claude 4.6/4.7 models default thinking to {"type": "adaptive"} when not explicitly provided. New tests add lightweight recording stubs to assert these parameter rewrites and preservation of existing output_config/extra_body fields.

^{Reviewed by Cursor Bugbot for commit 4241702. Bugbot is set up for automated code reviews on this repo. Configure here.}