Models & Labs
llama.cpp v0.3.0 Release
llama.cpp Releaseshigh confidence
Why it matters
- →The introduction of the dots3-note model expands multimodal capabilities in AI applications.
- →Improved tensor handling and MTP support enhance model performance and efficiency.
- →UI and server updates provide better usability and debugging options for developers.
llama.cpp has released version 0.3.0, featuring the new dots3-note multimodal model with a DSA-ISWA KV cache. This update includes MTP support for GLM-4.5-Air and improvements in tensor handling with ggml v0.22.0. The server now offers a new debug knob, and the web UI has been updated with tabbed chat navigation. These enhancements aim to improve performance and usability for developers working with multimodal AI applications.
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.
What happened next
- 1llama.cpp v0.4.0 Release
llama.cpp Releases · September 5, 2026 · Same story
- 2llama.cpp v0.4.1 Release Enhances Model Support
llama.cpp Releases · September 15, 2026 · Same story
Follow this story
Open the full story →Kimi-K3 Text Model Added to llama.cpp
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.
