What is LMSYS Org?

LMSYS Org (the Large Model Systems Organization) is a research organization that develops large models and systems described as open, accessible, and scalable. It is best known for two things: SGLang, a high-performance serving engine for large language and multimodal models, and Chatbot Arena, a crowdsourced benchmark where anonymous models compete in randomized battles. If you are looking for the team behind those projects, or for open tooling to train, serve, and evaluate LLM-based chatbots, this is the organization to know.

What LMSYS Org works on

The organization's stated focus is building large models and the systems around them — not just model weights, but the infrastructure for serving, training, and evaluating them. Its project list reflects that full pipeline:

A few things worth noting from this list:

• SGLang and FastChat are the two largest projects by stars (36.8k and 39.6k), and they cover different layers — SGLang focuses on high-performance serving, while FastChat spans training, finetuning, serving, and evaluation.
• Chatbot Arena is marked "Graduated," meaning it is no longer listed as an active project under the org's current project set.
• Vicuna's "90% ChatGPT quality"* is the organization's own characterization, with the asterisk indicating a qualified claim rather than a precise benchmark result.

Scale and community

The site reports the following figures:

• 15+ projects
• 87K+ GitHub stars
• 1000+ contributors

These numbers describe the breadth of the org's open-source footprint rather than any single product. The contributor count in particular suggests the projects are developed with substantial outside participation, which matters if you are considering adopting one of them — active external contribution usually means better odds of community support and continued maintenance.

Who backs it

LMSYS Org lists sponsors and partners including Voltage Park, NVIDIA, Nebius, Google Cloud, AtlasCloud, a16z, AMD, InnoMatrix, Laude Institute, Hyperbolic, NovitaAI, Verda Cloud, Sky9, Kaggle, MBZUAI, Together, RunPod, Anyscale, and HuggingFace. The site also has a "Sponsor Us" entry point.

The mix spans cloud/compute providers (NVIDIA, Google Cloud, AMD, Nebius, Voltage Park), AI infrastructure and serving companies (Together, Anyscale, RunPod, Hyperbolic, NovitaAI), and research institutions (MBZUAI, Laude Institute, Kaggle). That combination is consistent with an organization whose work depends on both GPU compute and deployment infrastructure.

When LMSYS Org is relevant to you

• You need to serve an LLM efficiently → look at SGLang.
• You want to train, finetune, serve, or evaluate a chatbot end to end → look at FastChat.
• You want to compare model quality through crowdsourced human preference → look at Chatbot Arena.
• You are researching LLM routing (deciding which model handles which query) → look at RouteLLM.
• You want an early open chatbot model → Vicuna is part of the org's history, though it is not presented as a current flagship.

Reading the org's blog

The site's "Latest Blog" section is where technical direction shows up first. Recent posts cover scaling decision models with SGLang, accelerating long-context and agentic inference with NVFP4 KV cache, and day-0 support in SGLang and Miles for DeepSeek-V4.1. If you are evaluating whether SGLang fits a specific workload — long context, agent sessions, or a newly released model — the blog is the place to check for current support rather than assuming it from the project description alone.

One caveat: the site does not publish pricing information, and nothing in the available material indicates whether any of these projects or services are free, paid, or gated behind login. Treat cost and access terms as something to verify per project.