v0.6.0 v0.7.0
ggml-org/llama.cppOct 5, 2026 → Oct 15, 202610 days between releases
What's New
Introduces speculative decoding, MoE GPU cache, and new models with breaking port/API changes
Breaking Changes
- Default listening port changed from 8080 to 9931.
- Chat API refactored around `common_chat_session` with `feed()` and `finish()` methods.
- Formatted chat output changed from a single string to an array of parts.
New Features
- LiliCorr speculative decoding technique for GLM5-Next models.
- GPU cache for MoE experts (`--moe-cache-mib`) offering speedups.
- Support for new models including K2 Horizon and EmbeddingGemma-2.
- Web UI Models Manager with search, filters, and sorting.
Improvements
- MoE expert cache shared across multiple GPUs.
- Optimized sparse attention path for MTP speculative decoding.
- Improved slot and prompt-cache restore behavior.
- ggml library updated to v0.27.0.
Upgrade Notes
Update your server configuration to listen on the new default port 9931. Refactor chat session code to use the new `common_chat_session` API methods `feed()` and `finish()`. Update chat output parsing logic to handle arrays of parts instead of a single string.
The Two Releases
v0.6.0
Oct 5, 2026· 16 days ago
llama.cpp v0.6.0 introduces a new extended batch API for mixed token/embedding inputs, adds support for the GLM-5.3-Flash (320B) and Clef decision models, and implements a new `/v1/systemone` API. It also includes Metal flash attention optimizations and a Web UI overhaul.
Full release notesv0.7.0
Oct 15, 2026· 6 days ago
llama.cpp v0.7.0 introduces speculative decoding techniques like LiliCorr and MTP for GLM5-Next, along with a new GPU cache for MoE experts that offers significant speedups. The release expands model support with new architectures including K2 Horizon and EmbeddingGemma-2, while updating the server API and adding a comprehensive models manager to the web UI.
Full release notes