Models & Labs
Llama.cpp b9291 Release Enhances MoE Prefill Throughput
llama.cpp Releaseshigh confidence
Why it matters
- →The update improves MoE prefill throughput, enhancing performance.
- →Expanded platform support increases accessibility for developers.
- →The focus on runtime efficiency makes llama.cpp more versatile.
The b9291 release of llama.cpp focuses on improving MoE prefill throughput by optimizing the SYCL implementation. This is achieved by changing the k_copy_src1_to_contiguous function to use a precomputed contiguous mapping, reducing complexity with a counting sort-based procedure. The update also broadens platform support, including macOS Apple Silicon and various Linux and Windows configurations. This release enhances llama.cpp's performance and compatibility, making it a more robust tool for developers.
Read originalThe story around this
Earlier coverage that leads up to this article, and what followed. Lines connect each piece to the closest one after it, converging here.
How we got here
- 1llama.cpp b9200 Release Enhances Performance
llama.cpp Releases · May 18, 2026 · Same story
What happened next
- 2llama.cpp b9290 Release Enhances SYCL Support
llama.cpp Releases · May 23, 2026 · Same story
More from llama.cpp Releases
Coding Toolscoding
llama.cpp b12022 adds ROCm 10 and Snapdragon support
This release quietly cements llama.cpp as the universal inference runtime by finally bringing ROCm 10.0 to Linux and Windows alongside CUDA 13.4, effectively closing the hardware gap for AMD users who were previously left behind. The inclusion of native Snapdragon support on Linux marks a significant step toward running local LLMs on ARM-based mobile chips without emulation overhead. While KleidiAI on Apple Silicon is disabled in this build, the broader expansion to diverse accelerators ensures that developers aren't locked into NVIDIA ecosystems anymore.
llama.cpp Releases
Coding Toolscoding
llama.cpp b12023 fixes shared prefix bug
This release addresses a specific inference artifact where decision prefixes were incorrectly bleeding into output embeddings. By capping shared prefixes at the first output position, llama.cpp ensures that option tokens remain distinct for parent and child tasks. This is a targeted fix for multi-step reasoning or voting workflows rather than a performance boost. It stabilizes complex generation patterns without changing the underlying model architecture.
