llama.cpp v0.6.0

Developers can use it to run supported models, build applications against its APIs, or serve models with multimodal and decision-model capabilities.

Availability

Released 05 Oct

Pricing model

Not established

Deployment

Builds listed for multiple operating systems and hardware backends

Privacy

Not established

Licence

Not established

• Linux and Windows builds are listed for CUDA 12 and CUDA 13.
• b11507 adds support for MoE cache over multiple GPUs. ^[1]

  llama: support MoE cache over multiple GPUs ( #30112 )

• The release lists macOS, iOS, Linux, Android, and Windows builds. ^[1]
• b11507 is marked as a pre-release. ^[1]
• The release extends Gemma-4 GPU decode through OpenCL to E2B. ^[2]

  opencl: enable Gemma-4 E4B GPU decode

• The release improves OpenCL support for FA and DK512 for Gemma-4. ^[2]
• The release lists builds for macOS, iOS, Linux, Android, Windows, and openEuler. ^[2]
• The release enables Gemma-4 E4B GPU decode through OpenCL. ^[2]

  opencl: enable Gemma-4 E4B GPU decode

• The release is marked Pre-release. ^[2]
• The release adds MTP support for Mellum2.1. ^[3]

  model : MTP support for Mellum2.1 ( #30612 )

• b11695 is marked as a pre-release. ^[3]
• Builds are listed for macOS/iOS, Linux, Android, and Windows. ^[3]
• The release adds inline RoPE construction in Mellum. ^[3]

  model : inline rope construction in Mellum

• The release adds MTP support for Mellum. ^[3]

  model : MTP support for Mellum2.1 ( #30612 )

• Paused downloads can be resumed, and incomplete downloads are remembered across restarts. ^[4]
• b11714 is marked as a pre-release. ^[4]
• The Discover Models surface includes a curated catalog list with search and a details pane. ^[4]

  The Discover surface: the curated-catalog list with its search and

• The Discover interface includes download options and a standalone download progress bar. ^[4]

  download options, and the standalone download progress bar that replaces

• The Discover setting gates browsing and downloading Hugging Face GGUF models. ^[4]

  The flag gates browsing and downloading HuggingFace GGUF models, so it is

• The new llama_batch_ext API supports mixed token and embedding batches and per-token state embeddings. ^[5]

  New llama_batch_ext extended batch API with llama_process() , supporting mixed token/embedding batches and per-token "state" embeddings for MTP and deepstack models #24669

• The server adds a /v1/systemone API for decision models. ^[5]

  llama and server: new /v1/systemone API supporting five decision models - laya, julia-1, lev, openjev (+vision), kev #29818

• The /v1/embeddings endpoint accepts typed vision, audio, and video content. ^[5]

  tools/server : new /v1/systemone endpoint for decision models #29818 and /v1/embeddings now accepts typed vision/audio/video content #29556

• The release adds support for the GLM-5.3-Flash 320B text-and-vision hybrid model and the Clef decision model with text and vision. ^[5]

  New models: GLM-5.3-Flash (GLM5-Next), a 320B text+vision hybrid model #27773 , and the Clef decision model, fully supported with both text and vision #29831 #29969

• llama.cpp v0.6.0 was released on 05 Oct. ^[5]

• DeepSeek adds web search for apps and updates its Flash model Daily brief 2026-10-17
• Generate editable slides and dashboards from a prompt Daily brief 2026-10-11