What projects does LMSYS Org develop?

Updated

LMSYS Org (Large Model Systems Organization) develops open, accessible, and scalable large models and systems. Its project portfolio spans LLM serving engines, chatbot training and evaluation platforms, and benchmarking tools — with 15+ projects, 87K+ GitHub stars, and 1000+ contributors as of the site's current listing. The main projects and their purposes are summarized below.

Core projects

Project Focus GitHub stars (as listed)
SGLang High-performance serving engine for large language models and multimodal models 36.8k
Chatbot Arena Benchmark platform for LLMs featuring anonymous, randomized battles in a crowdsourced manner Graduated
FastChat Open and scalable platform for training, finetuning, serving, and evaluating LLM-based chatbots 39.6k
Vicuna Open-source chatbot described as impressing GPT-4 with 90%* ChatGPT quality —
RouteLLM Framework for serving and evaluating LLM routers 5.6k

SGLang

SGLang is a high-performance serving engine for large language models and multimodal models. It is the project most directly tied to inference performance: recent LMSYS blog posts cover scaling decision models with SGLang, accelerating long-context and agentic inference with NVFP4 KV cache, and adding day-0 support for DeepSeek-V4.1 (including low-ratio compression and sliding-window attention with per-layer fp8 sliding-window cache). If your goal is deploying or optimizing model inference, SGLang is the relevant project.

Chatbot Arena

Chatbot Arena is a benchmark platform for LLMs that features anonymous, randomized battles in a crowdsourced manner. It is listed as "Graduated" in the current project listing. Its purpose is model comparison through human preference voting rather than static benchmarks.

FastChat

FastChat is an open and scalable platform for training, finetuning, serving, and evaluating LLM-based chatbots. It covers the full chatbot lifecycle, making it the entry point for teams that want to build and evaluate a chatbot rather than only serve a model.

Vicuna and RouteLLM

Vicuna is an open-source chatbot positioned around ChatGPT-quality conversation. RouteLLM is a framework for serving and evaluating LLM routers — useful when you want to route queries between models rather than commit to a single one.

How to choose

  • Serving or optimizing inference → SGLang.
  • Comparing models via human votes → Chatbot Arena.
  • Training, finetuning, or evaluating a chatbot → FastChat.
  • Routing between multiple models → RouteLLM.
  • A ready conversational model → Vicuna.

Sponsors and partners

LMSYS Org states it is backed by leading companies and institutions advancing AI research, listing 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 provides a "Sponsor Us" option.

For current project status, star counts, and new releases, check the Projects and Latest Blog sections on lmsys.org, since the figures above reflect the site's listing at the time of writing.