b10434
ggml-org/llama.cppb10434Aug 14, 2026by github-actions[bot]same day as b10433
What changed since b10433
b10434 adds OpenAI-style reasoning_effort plumbing into chat templates and generation params on top of b10433's ggml sync.
New features
- reasoning_effort from OpenAI Chat Completions requests is now stored in common_chat_templates_inputs and exposed to jinja chat templates.
- reasoning_effort is added to generation_params and placed alongside enable_thinking for consistent handling.
- Model-specific translations of reasoning effort are applied where a template requires a different naming or value scheme.
Improvements
- Server now correctly reads reasoning_effort from the request body, complementing the existing Responses API reasoning.effort conversion.
- Context is threaded through the analyze function and preserve_reasoning behavior is mirrored for consistency.
- Chat template inputs and generation parameters are better aligned with the OpenAI Chat Completions surface.
Bug fixes
- Fixed server-side parsing so reasoning effort supplied in the request body is actually propagated rather than dropped.
Upgrade notes
No configuration or model file changes are required; the new parameter is optional and defaults to prior behavior when absent. If you maintain custom jinja chat templates, you can now consume reasoning_effort to map effort levels to model-specific reasoning controls. Clients sending reasoning_effort via the OpenAI-compatible endpoint will see it honored instead of ignored.
Analyzed Oct 24, 2026
Upgrading?
Release notes summary
Adds support for OpenAI Chat Completions reasoning_effort parameter, making it available to jinja templates for better model control.
Key Highlights
- Added reasoning_effort parameter support
- Integration with OpenAI Chat Completions API
- Template compatibility improvements
- Generation parameter updates
Release Notes (excerpt)
<details open> chat : pass reasoning_effort to template * chat: add reasoning_effort to common_chat_templates_inputs Store OpenAI Chat Completions reasoning_effort and make it available to jinja templates (with model specific translations where required). Assisted-by: llama.cpp:Muse-Glimmer-30B * server : fixup reading reasoning effort from body server_chat_convert_responses_to_chatcmpl already handles conversion of Responses API reasoning.effort to reasoning_effort…
Published by ggml-org on GitHub.
Read the full release notesRelated Repositories
:electron: Build cross-platform desktop apps with JavaScript, HTML, and CSS
Godot Engine – Multi-platform 2D and 3D game engine
ClickHouse® is a real-time analytics database management system
Self-hosted game stream host for Moonlight.
DuckDB is a high-performance analytical database system