llama.cpp adds Qwen3-Next support; KataGo, Kimi-K3 models see updates

The latest release of llama.cpp (b10238) now supports Qwen3-Next with MTP, enabling optimized local inference on consumer hardware. Additionally, the Go AI engine KataGo has introduced an experimental evaluation cache in version v1.16.4 to speed up game state analysis. The Kimi-K3 multimodal model is also now accessible in the GGUF format, optimized by Unsloth for efficient local use on platforms like Hugging Face. AI

IMPACT Enhances local AI inference capabilities for developers and enthusiasts with improved model compatibility and performance optimizations.

RANK_REASON Updates to open-source AI inference engines and models.

Read on dev.to — LLM tag →

• Hugging Face
• KataGo
• Kimi-K3
• llama.cpp
• Qwen3-Next
• Unsloth

• model release
• infra

AI-generated summary · Google Gemini · from 1 sources. How we write summaries →