{"type":"post","id":"sglang-v0-5-23-ships-megakernel-low-latency-2254f8","slug":"sglang-v0-5-23-ships-megakernel-low-latency-2254f8","title":"SGLang v0.5.23 ships megakernel low-latency mode, Paged Experts for MoE on a single consumer GPU and TPU7x support","description":"SGLang v0.5.23 adds an opt-in path for running MoE models larger than GPU memory on one consumer GPU, alongside low-latency kernel modes and Google TPU7x support.","retrieval_nugget":"SGLang v0.5.23 adds an opt-in path for running MoE models larger than GPU memory on one consumer GPU, alongside low-latency kernel modes and Google TPU7x support. The release matters to teams operating their own model-serving stack because it expands deployment options and reports substantial performance gains, though those claims remain workload- and hardware-specific.","published_at":"2026-11-03","updated_at":"2026-11-03","record_date":"2026-11-03","date_kind":"published_at","topics":["google"],"entities":["SGLang","SGLang-TorchTPU","DeepSeek-V4.1","DeepSeek-V4.1-Flash","Google TPU7x","GLM-5.3-Flash","gpt-oss"],"related_patterns":["harness-architecture-outlives-model-choice"],"novelty_label":"new_fact","what_changed":"SGLang v0.5.23 adds megakernel low-latency mode, Paged Experts for MoE on a consumer GPU, and TPU7x support—new capabilities beyond v0.5.22 and the earlier TPU engine announcement.","prior_refs":["https://newruntime.com/posts/announcing-new-vllm-and-sglang-inference-engines-on-be000f/"],"source_url":"https://github.com/sgl-project/sglang/releases/tag/v0.5.23","source_urls":["https://github.com/sgl-project/sglang/releases/tag/v0.5.23"],"schema_version":"newruntime-agent-readable-v0.2","stable_id":"post:sglang-v0-5-23-ships-megakernel-low-latency-2254f8","status":"published","visuals":[],"editor":{"role":"editor","name":"Andrey Reshetnikov","profile":"https://newruntime.com/owner-profile.md","profile_json":"https://newruntime.com/owner-profile.json","same_as":["https://www.linkedin.com/in/reshetnikov1/","https://t.me/lapetuse"]},"routes":{"html":"https://newruntime.com/posts/sglang-v0-5-23-ships-megakernel-low-latency-2254f8/","markdown":"https://newruntime.com/posts/sglang-v0-5-23-ships-megakernel-low-latency-2254f8.md","json":"https://newruntime.com/posts/sglang-v0-5-23-ships-megakernel-low-latency-2254f8.json"},"markdown_url":"https://newruntime.com/posts/sglang-v0-5-23-ships-megakernel-low-latency-2254f8.md","next_reads":[{"type":"topic","path":"/topics/google/","url":"https://newruntime.com/topics/google/","title":"Google","reason":"Explore the google topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/google.md","json_url":"https://newruntime.com/topics/google.json"},{"type":"collection","path":"/posts/","url":"https://newruntime.com/posts/","title":"Field Notes","reason":"Open the latest editorial analysis.","media_type":"text/html","json_url":"https://newruntime.com/posts.json"},{"type":"collection","path":"/sources/","url":"https://newruntime.com/sources/","title":"Sources","reason":"See the original sources this archive cites.","media_type":"text/html"}]}