Clarity for AI buildersSponsored by AI Routing
availability

LMSYS launches SGLang Consortium with eight founding members to coordinate open-source inference work

LMSYS launches SGLang Consortium with eight founding members to coordinate open-source inference work
LMSYS Org has launched the SGLang Consortium to coordinate engineering work on open-source AI inference. In its November 12, 2026 announcement, LMSYS named NVIDIA, AMD, Intel, Google, Meta, Cloudflare, Modal and RadixArk as founding members. The nonprofit announced the group at the opening keynote of the inaugural SGLang Summit in San Francisco. Members will work with SGLang’s maintainers on frontier model support, reasoning and multimodal workloads, distributed inference and hardware optimization. Each founding member will designate a technical representative and participate in two roadmap meetings a year. Contributions can include engineering expertise, deployment feedback, testing resources or joint development. Working groups may collaborate between meetings, and the announcement calls for public updates on roadmap direction and progress. The arrangement creates a formal channel for participating companies to influence priorities, but it does not transfer control of the project to them. LMSYS says SGLang’s maintainers retain responsibility for technical decisions and code acceptance, and the project remains open to the wider community. The distinction is between coordinating engineering needs and deciding which changes enter the codebase. LMSYS describes SGLang as a full-stack inference engine in the PyTorch ecosystem and says it runs on hundreds of thousands of GPUs. It names Meta, Microsoft, Google, Amazon, Thinking Machines Lab, Cloudflare, Reflection and Modal among companies whose production inference deployments it powers. These deployment claims come from the announcement; the supplied material does not include independent usage measurements. The proposed contribution model connects production experience with implementation work: deployment feedback can identify needs, while engineering and testing resources can support changes intended to address them. That is the practical mechanism described for the consortium, rather than a newly released software capability. The announcement does not specify individual engineering deliverables, release dates or measured performance improvements resulting from the collaboration. The announcement also cites Deloitte’s expectation that inference will account for roughly two-thirds of AI compute in 2026, up from about half in 2025. That forecast provides context for the focus on inference efficiency, but it is not evidence that the consortium has already reduced costs or increased throughput. Ying Sheng, a co-author of the original SGLang paper and RadixArk’s co-founder and CEO, framed software efficiency as important amid challenges in building new data centers. The launch took place at the first SGLang Summit, held November 12–13 at San Francisco’s Fort Mason Center for Arts & Culture. According to LMSYS, more than 100 speakers are taking part, with sessions spanning reinforcement learning, multimodal models and hardware platforms.