You signed in with another tab or window. Reload to refresh your session.You signed out in another tab or window. Reload to refresh your session.You switched accounts on another tab or window. Reload to refresh your session.Dismiss alert
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_tokens with thinking.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.