One decision model sees and hears
Cloudflare released Clef-omni on October 9 as an open-weight decision model that accepts audio and video alongside text and images. The company’s example asks whether an appliance label is visible, whether its sound seems normal and whether a fan is running. Those checks combine a photo, audio recording and video clip in one structured request.
Cloudflare says a 21-second video with sound can be scored in about 1.5 seconds through a single API call. Because Clef models return constrained decisions rather than a stream of generated text, the company says it can skip a separate speech-to-text step and synchronize audio with visual frames directly. The weights are based on Qwen3-Omni-30B-A3B-Instruct, with the speech-generation components removed.
Why this is a meaningful update
Many multimodal systems assemble several components: transcribe audio, inspect images, then ask a language model to combine the results. Clef-omni is designed to map evidence from different modalities into structured choices. That could simplify workflows such as equipment inspection, form processing or reviewing media against a checklist.
The family has limits. Cloudflare’s own benchmark table shows Clef-omni does not lead Clef on every task; for example, its appliance exact-match result is lower. The right question is whether a single fast call works well enough for a particular decision, not whether it replaces general-purpose models.
The pricing trade-off to know
Cloudflare also cut Clef-flash’s hosted input price from $0.09 to $0.038 per million tokens, while reducing its hosted context window from 64K to 24K. The open weights remain trained to support a larger context when self-hosted, according to the company. Clef-omni launched at $0.15 per million input tokens.
That makes the family easier to evaluate across cost, speed and context needs. If a workflow needs only short structured judgments, the lower-cost option may fit. If it needs a long input, the hosted context change matters. These are new capabilities and pricing from a just-updated family, building on the earlier Clef release.
Source published 2026-10-09. Coverage is based on the maker’s announcement and demonstration.