LMSYS Org announced the SGLang Consortium at the inaugural SGLang Summit in San Francisco on November 12. The founding members are NVIDIA, AMD, Intel, Google, Meta, Cloudflare, Modal and RadixArk, bringing together rival hardware makers and large inference users around a shared open-source software project.
SGLang is an inference engine used to serve language models, including reasoning and multimodal workloads. Inference software affects how efficiently deployed hardware turns model requests into responses, making it an important layer of the AI infrastructure stack.
The consortium is designed as a coordination forum rather than a transfer of control over the project. Each founding member will designate a technical representative and participate in two roadmap meetings annually, with possible working groups for additional collaboration.
The stated agenda includes support for frontier models, reasoning, multimodal inference, distributed serving and hardware optimization. These are proposed workstreams; the launch itself does not establish new benchmark records or finished engineering improvements.
The significance is the breadth of competing vendors committing to work with the same open serving stack. Cross-platform software can reduce lock-in and potentially improve efficiency, but actual benefits will depend on subsequent code contributions, tests and adoption.