framework/v1.9.0 plugins/compat/v0.4.0
maximhq/bifrostOct 13, 2026 → Oct 13, 20260 days between releases
What's New
Upgraded compat plugin adds reasoning-only model toggles and skips native provider emulation
New Features
- Added `convert_decision_to_responses` setting to control decision emulation behavior.
- Added `force_reasoning_only_models_to_responses` toggle to force reasoning-only models to responses.
- Skip decision emulation for providers served natively by OpenAI, Ollama, and others.
- Report `reasoning.mode` as dropped on providers without native support.
Improvements
- Upgraded core to v1.12.0 and framework to v1.9.0.
- Internal refactor: `HTTPTransportResponseHeadersHook` is now a no-op.
- Improved token budget fitting via `BifrostContextKeyConvertParamValues`.
Bug Fixes
- Drop `reasoning.summary` for models that do not support it.
- Fix output token and thinking budget fitting to core per attempt.
Upgrade Notes
The compat plugin now introduces `convert_decision_to_responses` and `force_reasoning_only_models_to_responses` settings. Note that `force_reasoning_only_models_to_responses` defaults to `true`, which may alter reasoning handling for supported providers. Ensure your configuration includes these new settings to maintain your desired provider behavior.
The Two Releases
Framework v1.9.0
Oct 13, 2026· 9 days ago
This release focuses on extensive persistence improvements for agents and logs, introduces the new Warp agent for semantic search, and adds infrastructure enhancements like a multi-backend queue system and improved MCP integrations.
Full release notesPlugin compat v0.4.0
Oct 13, 2026· 9 days ago
The compat plugin introduces new configuration settings to control decision emulation and reasoning handling for specific providers like OpenAI and Ollama.
Full release notes