SGLang 0.5.15 Boosts Inference Speed
BanghuaZ · x · 2026-07-12
SGLang v0.5.15 has been released, focusing on performance optimizations for production inference.
- Official benchmarks indicate that on 8x B300, production serving for GLM-5.2 NVFP4 achieves 500+ tok/s/user; on 4x GB300, it reaches 450 tok/s/user (bs=1).
- This release cycle primarily focused on inference stack tuning, with full technical details and run commands coming soon.
- Newly supported models include: Hunyuan 3 (Hy3), HRM-Text, NVIDIA LocateAnything-3B, Baidu Unlimited-OCR, JoyEcho, and Qwen3.6.
- Version highlights also feature:
- Breakable CUDA Graph becoming the default capture path
- Built-in web search powered by Exa
- decode context parallelism for MLA models, including DeepSeek V3
- FlashInfer all-to-all for routing models
Related event: SGLang v0.5.15 Boosts Production Inference Performance(2 posts)→
More from Infra
- Two CVSS 9.9 Wasmtime flaws allow sandbox escape, one ARM64-only since 2020 — basedjensen · 2026-10-27
- AI Infra Meme: While Countries Need TW-Scale Power for Datacenters, People Argue About Lightbulbs — mgostIH · 2026-10-27
- Legora runs 2,000 parallel agent evals on Modal Sandboxes for legal AI — blelbach · 2026-10-27
- Cloud release velocity scores: AWS 95, Azure 90, Google Cloud 82 — DavidLinthicum · 2026-10-27
- Selling a 5090 to scale local inference: one business weighs 5 hardware paths — Anarchaotic · 2026-10-27
- AWS tutorial: fleet-wide edge CV model lifecycle with SageMaker SDK v3, ~$0.07 per run — AWS ML Blog · 2026-10-27