AI Trends

Trending story

SGLang Consortium Established with Nine Founding Members Joining

1 reports1 story updatesUpdated Nov 17 08:00

Understand this story

AI overview

On November 12–13, 2026, LMSYS hosted the inaugural SGLang Summit at Fort Mason in San Francisco, with more than 2,000 developers in attendance. At the event, the establishment of the SGLang Consortium was announced, with nine founding members: NVIDIA, AMD, Intel, Google, Meta, Thinking Machines Lab, Cloudflare, Modal, and RadixArk. The same event also released SGLang v1.0 and Miles v0.2: SGLang v1.0 focuses on the efficiency and scalability of production inference, addressing the state management needs of hybrid-architecture models and long-running agents through unified memory management and tiered caching; Miles v0.2 extends reinforcement learning post-training to NVIDIA GPUs, AMD GPUs, and Google TPUs, and supports images, video, audio, and diffusion models. Meta stated that its production inference is already powered by SGLang, and that migrating to it has saved more than 2x in GPU costs.

AI-generated from public reports

Latest development · Report

SGLang First Summit Recap: Nine-Member Consortium Formed, SGLang v1.0 and Miles v0.2 Released

Developments

1 confirmed developments

  1. SGLang Consortium established, nine founding members join

    Source:LMSYS Blog

Reports and discussion timeline

Browse reports and discussion by time. Their evidence levels remain distinct.

Nov 17

  1. LMSYS BlogFirst-partySelected

    SGLang First Summit Recap: Nine-Member Consortium Formed, SGLang v1.0 and Miles v0.2 Released

    LMSYS held the first SGLang Summit on November 12–13 at Fort Mason in San Francisco, with more than 2,000 developers attending. At the event, it announced the formation of the SGLang Consortium, composed of nine founding members: NVIDIA, AMD, Intel, Google, Meta, Thinking Machines Lab, Cloudflare, Modal, and RadixArk, and released SGLang v1.0 and Miles v0.2. SGLang v1.0 focuses on the efficiency and scalability of production inference, using unified memory management and tiered caching to address the state management needs of hybrid-architecture models and long-running agents; Miles v0.2 extends reinforcement learning post-training to NVIDIA GPUs, AMD GPUs, and Google TPUs, and supports images, video, audio, and diffusion models. Meta said its production inference is already powered by SGLang, and that GPU costs were reduced by more than 2x after migration.

Related stories

  1. Related storiesQwen Releases Qwen4-Max Preview Version
  2. Related storiesGoogle launches Gemini 4 Hydrogen on Antigravity
  3. Related storiesRecursive releases open-weight local coding model Flint
  4. Related storiesMeta releases open-weight model Muse Spark Open