New in Llama.cpp: Anthropic Messages API (huggingface.co)

The latest update to the llama.cpp server introduces support for the Anthropic Messages API, enabling users to leverage Claude-compatible clients with locally-running models. This feature addresses a strong demand within the community for compatibility with tools like Claude Code, allowing for smoother interactions with local AI models. The implementation translates Anthropic's data format to OpenAI's format, ensuring seamless integration with the existing inference pipeline. This advancement is significant for AI/ML practitioners as it enhances the flexibility and capability of local model deployment, particularly with tools requiring intelligence processing, such as coding assistants. The full Messages API includes functionality for chat completions, token counting, function calling, and even image input for multimodal models. Users can initiate the API with simple curl commands, allowing for efficient experimentation and integration in their applications. This update not only streamlines usage for existing llama-server users but also showcases the potential for running specialized models like Qwen3 Coder and MiniMax M2 to optimize agentic workloads, marking a notable step forward in AI local processing capabilities.