Brief
llama.cpp b10996 forces a newline at the end of the reasoning budget for qwen3-coder in chat
The llama.cpp release b10996 carries one behavioural change: chat now forces a \n at the end of the reasoning budget for qwen3-coder. The release note gives the commit title and the build matrix, and nothing more, so the mechanism and the user-visible effect are not documented in the evidence.
Release b10996 of llama.cpp lists a single chat change: forcing a newline when the reasoning budget ends for qwen3-coder, referencing pull request 28869. The same page enumerates the downloadable builds — macOS Apple Silicon (arm64), Ubuntu, Android and Windows images across CPU, CUDA, Vulkan, ROCm, OpenVINO and SYCL backends — with macOS Apple Silicon (arm64, KleidiAI enabled) and openEuler marked DISABLED.
For anyone running qwen3-coder through llama.cpp's chat path, this is the kind of small formatting fix that can alter how a truncated or budget-limited reasoning turn is terminated. That is the whole of the published detail: no rationale, no before/after example, no note on which other models or interfaces are affected, and no indication of whether the newline is inserted into the stored transcript or only the output stream.
Our reading
This matters to people who self-host qwen3-coder behind llama.cpp's chat interface or a server built on it, because their downstream parsing or prompt-continuation logic may key off exactly where a reasoning block ends. It does not matter to API users of hosted qwen3-coder, who are not touched by this binary. Because the note is a commit title, the safest reading is that it is a targeted fix rath…
What to do or watch
If you run qwen3-coder locally through llama.cpp chat, update to b10996 and compare one reasoning-heavy prompt against your previous build, watching specifically for trailing-newline differences in the output you parse. The unresolved question is whether this newline appears in the stored transcript or only in the streamed output — the release note does not say.
Source details and supporting facts
Each line is stated by the page named above it.
Stated by github.com
- Release b10996 of llama.cpp includes a chat change described as forcing a newline on reasoning budget end for qwen3-coder, referencing #28869.
- The release page lists builds for macOS (Apple Silicon arm64, Apple Silicon arm64 with KleidiAI enabled DISABLED, Intel x64, iOS XCFramework), Linux (Ubuntu x64 and arm64 CPU, s390x, Vulkan, CUDA 12 and 13, ROCm 10.0, OpenVINO, SYCL FP32 and FP16), Android arm64, Windows (x64 and arm64 CPU, arm64 O…
Sources
- llama.cpp releases (GitHub)Text stored 16 September 2026
How this story was checked. Written from the 1 page listed above, stored 16 September 2026; claims checked against that stored text on 16 September 2026.
What that means
- 2 of 3 reported statements were confirmed against the page that carries them; the rest were removed rather than published.
- Figures in the text were required to appear in the stored source text: yes. Identifiers: yes.
- The check reads stored text only: no claim rests on a fresh look that did not happen.
- Where the reporting was silent, the text says so instead of filling the gap.