SGLang v0.5.19 release: new models and beam search
Melvin Vivas · X post · 2026-09-10 · Open on X
Topics: LLMOps, Deployment & Monitoring, Industry Trends & Job Market · Level: intermediate
Summary
Announces SGLang v0.5.19, an open-source LLM serving and inference engine. The release has 786 PRs from 214 contributors, 51 of them new. It adds support for several new open models plus LongCat-Image-Edit on SGLang-Diffusion, and the quoted post says beam search has been updated.
Key points
- SGLang v0.5.19 has 786 PRs from 214 contributors, 51 of them first-time contributors.
- New model support: Qwen3.8 and Qwen3.8-27B, dots3.note, Ling-3.0-flash, Ling-3.0-tiny, Spark2.5 and Granite 4.2.
- SGLang-Diffusion now supports LongCat-Image-Edit for image editing.
- The quoted post says beam search has changed, but the text is cut off before the details.
- SGLang is an option for self-hosting open-weight models in production, like other serving engines.
Resources mentioned
- SGLang · repo · github.com · free
Open-source, high-performance serving engine for running LLMs and multimodal models. - SGLang-Diffusion · tool · docs.sglang.io · free
SGLang's add-on for serving diffusion and image-generation models. - Qwen3.8 · tool · qwen.ai · free
Upcoming large Qwen model (2.4T parameters) that Qwen says will be released with open weights.
Also in: Qwen3.8-Max on the Frontend Code Arena cost-performance frontier (Melvin Vivas on X · notes), Qwen3.8-Max announced with open weights for Max and 27B (Melvin Vivas on X · notes), Qwen3.8 (2.4T params) Announced as Upcoming Open-Weight Model (Melvin Vivas on X · notes), Qwen3.8 Open-Weight Launch Announcement Reaction (Melvin Vivas on X · notes) - Granite 4.2 · tool · research.ibm.com · free
IBM's open Granite language model. - Ling-3.0-flash / Ling-3.0-tiny · tool · huggingface.co · free
Ling 3.0 family of open language models in flash and tiny sizes. - dots3.note · tool · huggingface.co · free
Open model newly supported in SGLang. - Spark2.5 · tool · check price
Model newly supported in SGLang. - LongCat-Image-Edit · tool · github.com · free
Image-editing diffusion model served through SGLang-Diffusion.
Try this
- Upgrade to SGLang v0.5.19 to serve the newly supported models.
More in LLMOps, Deployment & Monitoring
- llama.cpp v0.4.1 release announcement
- DeepSeek V4.1 Flash Off-Peak Pricing as a Cheap Fallback Model
- How inference engines work: the full life of an LLM request
- llama.cpp's built-in web UI for testing local models
- Local Speech-to-Text with LFM2.5-Audio-1.5B and llama.cpp
- Daytona Offers GPU Sandboxes (H100, RTX 4090/5090)