Cloudflare’s Clef-omni can score text, images, audio and video in a single decision call, according to MarkTechPost. The model could spare developers some transcription and captioning work, but the reported tests also show it is not a reasoning upgrade across the board.
Cloudflare Watch analysis
What happened
MarkTechPost says Clef-omni is an open-weight, 30-billion-parameter model with 3 billion active parameters, built from Qwen3-Omni. The report says it can take synchronised audio and video alongside text and images, then return scored options rather than generated prose. It describes support for up to four images, four audio clips and two video clips per request, with a 64,000-token context window on Workers AI.
The report attributes benchmark results to Cloudflare. Clef-omni led the compared models on two intent-classification tests, with scores of 94.8 macro-F1 on BANKING77 and 97.7 on CLINC150+OOS. But it trailed text-focused Clef and Jev on several reasoning and workflow tests. MarkTechPost also says Cloudflare’s hosted Clef service became faster after a move to SGLang, while Clef-flash pricing was cut. The article gives specific speed comparisons but no exact new Clef-flash price in its summary.
Why it matters
For developers building systems that must make decisions from video or sound, handling several input types in one call may simplify the pipeline. That is useful engineering, not a universal intelligence upgrade: the reported results point to strength in classification and weaker performance on some harder reasoning tasks.
The distinction matters when choosing a model for a real workflow. A system sorting customer requests may benefit from quick, structured decisions; one expected to reason through a complicated case may need a different tool or human review.
Our read
Clef-omni has a clear job description: make structured decisions across more kinds of input. That is more interesting than another model claiming to do everything, and the uneven benchmark results are part of the useful story, not a footnote to hide under the carpet. Treat the scores as results reported from Cloudflare’s comparisons, and test the model on the task you actually need it to do.
What to watch
- Whether developers can use the multimodal inputs reliably in deployed workflows.
- How Clef-omni performs on independent evaluations, especially reasoning tasks.
- Whether Cloudflare publishes clearer pricing comparisons for Clef-flash.
Discussion spark: Would you use one model for decisions across text, images, audio and video if it simplified your pipeline, or keep separate models for each input type?
Sources and evidence
- Meet Cloudflare Clef-omni: A 30B-A3B Multimodal Decision Model for Audio, Video, Image and Text – MarkTechPost (11 October 2026, 23:15 UTC)
not affiliated with or endorsed by Cloudflare