SGLang announces that Cloudflare now serves the Clef model through SGLang's existing /v1/systemone API. The team says it contributed Clef support and performance optimizations upstream, validated against the official reference implementation.
Clef-Flash reaches sub-40 ms decision latency on a single H200, answers 64 questions in one prefill with text and image support, and Clef-Omni support is coming next.
The post names the specific serving path, latency measurement, and hardware used, so readers can judge how Clef runs on SGLang.
Very proud to see @Cloudflare serving Clef with SGLang! Now available through SGLang’s existing /v1/systemone API.
The team also contributed Clef support and performance optimizations upstream to SGLang with:
· Validated against the official reference implementation
· Sub-40 ms decision latency for Clef-Flash on a single H200
· 64 questions answered in one prefill, with text and image support
Clef-Omni support is up next!
we've got more 𝚌𝚕𝚎𝚏 for you: • new 𝚌𝚕𝚎𝚏-𝚘𝚖𝚗𝚒 - takes audio, video, image, and text • 𝚌𝚕𝚎𝚏-𝚏𝚕𝚊𝚜𝚑 is now cheaper than jev • 𝚌𝚕𝚎𝚏 is now ~2x faster https://blog.cloudflare.com/clef-faster-cheaper-multimodal/View quoted post on X
Source: SGLang · x.comPublished · added here
