The development

What happened

Cloudflare expanded its open-weight decision-model family on 9 October with Clef-omni, adding audio and video to text and image inputs. The release also changes the cost and capacity of hosted Clef-flash and speeds up hosted Clef. It gives developers distinct options for multimodal decisions, cheaper short requests and larger contexts.

Key facts

  • Clef-omni accepts WAV/MP3 audio and MP4/WebM video alongside text and images in one schema-constrained decision request, rather than generating a conversational response.

  • Announced input-token rates per million are $0.15 for Clef-omni, $0.038 for Clef-flash (previously $0.09), and $0.24 for Clef. Media-token conversion affects the billed input size.

  • Hosted Clef-flash drops from a 64k to a 24k context window. Its released weights are unchanged and were trained for 256k context; hosted Clef remains at 64k.

  • Cloudflare reports Clef median latency falling from 616 to 305 milliseconds at about 3,400 input tokens. Its other tested sizes improved by 1.7 times; the 2-times result is workload-specific.

  • The Clef speed changes come from serving infrastructure, with no new weights. Clef-omni's vendor benchmark results vary by task and do not beat every listed comparator.

The context

Why it matters

TideReach assessment: a single multimodal decision request could simplify workflows that otherwise transcribe audio or separate video into several processing stages. The price cut also creates a practical migration question: a workload that relied on Clef-flash's larger hosted context may need shorter inputs or another model. Buyers should compare representative input lengths, media charges, decision accuracy and tail latency together. Vendor medians alone cannot establish reliability or total cost in a production application; TideReach has not independently benchmarked these models.