llama.cpp Release b11136 Adds Standard OpenAI Video API Support
Release b11136 of llama.cpp updates `llama-server` to support the standard OpenAI `video_url` content type and `data:video/*` URIs. This fixes issues where OpenAI-compatible clients failed when sending base64-encoded video data or standard video payload structures. Improving OpenAI API specification compliance ensures seamless interoperability with third-party tools, libraries, and frontends designed for OpenAI's multimodal models. Developers using `llama-server` as a local drop-in replacement can now process video inputs without needing custom client-side workarounds. The patch configures `llama-server` to treat `video_url` as an alias for `input_video` and extract media objects regardless of which key is provided. Additionally, it enables support for `data:video/*` URIs, matching the server's existing capability for handling image data URIs.
## BACKGROUND
`llama.cpp` is a widely used open-source C/C++ framework for running large language models locally across various hardware backends. Its built-in HTTP server component, `llama-server`, exposes an OpenAI-compatible endpoint so applications can seamlessly switch between cloud LLM services and locally hosted models.