Changes
PR #2270 by pcbeingused333 and kaylvan: fix(anthropic): send adaptive thinking and no sampling params to Haiku 5.5, Mythos and Opus 4.7/4.8
- Fixed Anthropic requests by omitting sampling parameters (such as the default 0.7 temperature) for Claude Haiku 5.5 and Mythos models, which reject non-default values.
- Extended adaptive thinking support to Opus 4.7, Opus 4.8, Haiku 5.5, and Mythos 5/5.1, preventing 400 errors caused by sending
budget_tokenswiththinking.type=enabled; Opus 4.6 remains the newest model that accepts it. - Added new Claude models, including Fable 5.1, Sonnet 5.5, Haiku 5.5, and Mythos 5 and 5.1.
- Enabled million-token context beta support for all newly listed models.
- Upgraded dependencies: mimetype to v1.4.16, Ollama to v0.40.6, and openai-go to v3.81.0.