SGLang

license

Apache-2.0(OSI)

source

public

core-gated

ungated(Apache-2.0 across the repo, no ee/ or enterprise directory among its top-level directories, and no paid tier that withholds features

governance

LMSYS/PyTorch-ecosystem(community)

Apache-2.0 throughout and no functionality withheld from the published source. The repo carries no ee/ or enterprise directory, no license-key gate and no closed dependency; what SGLang offers commercially is consulting and sponsorship, which software.yaml explicitly does not count as gating a core. core-gated is recorded ungated.

• https://github.com/sgl-project/sglang recorded 2026-08-11

  Repo page for SGLang, "a high-performance serving framework for large language models and multimodal models". Licensed Apache-2.0. Top-level directories are .claude, .devcontainer, .github, 3rdparty/amd, assets, benchmark, docker, docs, examples, experimental/sgl-router, proto, python, rust, scripts, sgl-model-gateway, test -- no ee/ or enterprise directory. Nothing on the page names a closed package, a license key, or a paid tier; the only commercial line is an invitation to "enterprises interested in adopting SGLang at scale, including technical consulting, sponsorship opportunities".

• https://raw.githubusercontent.com/sgl-project/sglang/main/LICENSE recorded 2026-08-11

  Apache License Version 2.0, January 2004

GitHub repo states SGLang "has become the de facto industry standard, with deployments running on over 400,000 GPUs worldwide." A third-party guide corroborates: deployed on 400,000+ GPUs generating trillions of tokens/day, with xAI, NVIDIA, AMD, LinkedIn, Cursor, and Oracle/Google/Azure/AWS clouds named as users. Level 4 reflects a very large but operator-concentrated deployment footprint (no public DAU/install-base figure).

• https://github.com/sgl-project/sglang recorded 2026-08-09

  SGLang has become the de facto industry standard, with deployments running on over 400,000 GPUs worldwide.

• https://inference.net/content/sglang-complete-guide/ recorded 2026-08-09

  SGLang is deployed on over 400,000 GPUs worldwide, generating trillions of tokens daily. Major technology companies have built their inference infrastructure on SGLang: xAI uses SGLang for Grok inference

An independent H100 benchmark on Llama-3.3-70B at FP8 on a single GPU recorded SGLang at 1,920 tokens per second and 112ms to first token, against vLLM's 1,850 tokens per second and 120ms, so SGLang matched or slightly exceeded vLLM on that workload, while the same write-up frames vLLM as the broader-purpose default and SGLang as strongest for shared-prefix workloads. The two therefore sit level with each other. Neither the project's GitHub page nor that write-up records an SGLang MLPerf submission, so the score rests on the feature set, RadixAttention prefix-KV reuse, a zero-overhead CPU scheduler, prefill-decode disaggregation, speculative decoding, structured outputs and hardware support across NVIDIA, AMD, Intel, TPU and Ascend, together with third-party benchmarks, rather than on a standardized suite.

• https://www.spheron.network/blog/vllm-vs-tensorrt-llm-vs-sglang-benchmarks/ recorded 2026-08-09

  vLLM ... 1,850 tok/s 120 ms ... SGLang Shared-prefix workloads, low latency 1,920 tok/s 112 ms

• https://github.com/sgl-project/sglang recorded 2026-08-09

  Fast Runtime: Provides efficient serving with RadixAttention for prefix caching, a zero-overhead CPU scheduler, prefill-decode disaggregation, speculative decoding, continuous batching, paged attention