{"type":"analysis","id":"analysis:research-vllm-and-sglang-can-match-deepinfer-per-kernel-step-but-their-benchmarks-don-t","slug":"research-vllm-and-sglang-can-match-deepinfer-per-kernel-step-but-their-benchmarks-don-t","title":"vLLM and SGLang can match DeepInfer per kernel step, but their benchmarks don't show production parity","description":"SGLang's decode test lines up closely with DeepInfer's own benchmark. vLLM's comparison spans different hardware and setups, and its revenue-per-GPU figure comes from a different workload. Nobody has compared full serving stacks under a stated latency target.","published_at":"2026-11-09T02:00:00.000Z","updated_at":"2026-11-09T02:00:00.000Z","record_date":"2026-11-09","topics":["benchmarks","open_source","research"],"source_urls":["https://github.com/deepseek-ai/DeepInfer","https://vllm.ai/blog/2026-10-23-deepseek-v41-flash-pd-wide-ep","https://www.lmsys.org/blog/2026-10-24-dsv41-deepinfer/","https://github.com/deepseek-ai/DeepInfer/blob/main/doc/inference_deployment_qa.md","https://the-decoder.com/deepseek-open-sources-the-inference-engine-that-powers-its-fastest-and-cheapest-ai-model/","https://cellcog.ai/blog/deepseek-deepinfer-open-source"],"language":"en","routes":{"html":"https://newruntime.com/analysis/research-vllm-and-sglang-can-match-deepinfer-per-kernel-step-but-their-benchmarks-don-t/","markdown":"https://newruntime.com/analysis/research-vllm-and-sglang-can-match-deepinfer-per-kernel-step-but-their-benchmarks-don-t.md","json":"https://newruntime.com/analysis/research-vllm-and-sglang-can-match-deepinfer-per-kernel-step-but-their-benchmarks-don-t.json"},"author":{"role":"author","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"]},"editorial_provenance":{"desk":"authorial","target_ref":"deep_research:09bb2154-e95f-4ddf-9806-0a806aa8b454","authorial_article_id":"09bb2154-e95f-4ddf-9806-0a806aa8b454"},"next_reads":[{"type":"topic","path":"/topics/open-source/","url":"https://newruntime.com/topics/open-source/","title":"Open Source","reason":"Explore the open_source topic hub.","media_type":"text/html","markdown_url":"https://newruntime.com/topics/open-source.md","json_url":"https://newruntime.com/topics/open-source.json"},{"type":"collection","path":"/analysis/","url":"https://newruntime.com/analysis/","title":"Analysis","reason":"Open the latest authorial analysis.","media_type":"text/html","json_url":"https://newruntime.com/analysis.json"},{"type":"collection","path":"/sources/","url":"https://newruntime.com/sources/","title":"Sources","reason":"See the original sources this archive cites.","media_type":"text/html"}]}