正文 · AI 翻译
IT Home reported on October 10 that Cloudflare issued an announcement yesterday (October 9) introducing Clef-omni, an open-weight decision model that can process text, images, audio, video and more in a single API call.
In terms of its positioning, the model builds on the earlier Clef series by adding support for multimodal inputs. Its model weights are available on Hugging Face.
As for multimodal support, the earlier Clef supported text, images and sequences of frames extracted from video (extracting still images from a video over time, then feeding them to the model in order). Clef-omni adds audio and full video input, supporting wav, mp3, mp4 and webm formats.
Cloudflare says developers can use one model to handle multiple types of input without separately setting up speech transcription and audio-video splitting workflows.
Cloudflare says Clef-omni is built on Qwen3-Omni-30B-A3B-Instruct and retains its main comprehension capabilities. The model is designed for structured decision tasks and does not generate conventional text output. In tests published by the company, the median response time for text-only requests was about 130 milliseconds, compared with about 150 milliseconds for images; it scored a 21-second video with sound in about 1.5 seconds.
BenchmarkClef-omniClefClef-flashJev
BFCL: exact-match rate98.298.4798.7695.75
ToolRet: nDCG@1066.669.1966.4365.28
API-Bank: accuracy92.791.9393.1188.19
Household appliances: exact-match rate69.382.9597.7352.27
When2Call: accuracy63.372.3765.5880.97
BANKING77: macro-average F194.894.2090.9379.74
CLINC150+OOS: macro-average F197.797.4366.7789.27
BRIGHT: nDCG@1042.045.9139.2647.52
Amazon ESCI: macro-average F157.857.4857.3955.21
PhishNChips: accuracy73.279.6075.0562.55
Cloudflare also cut the price of Clef-flash per million input tokens from $0.09 (IT Home note: approximately 0.6 yuan at the current exchange rate) to $0.038 (approximately 0.25 yuan at the current exchange rate), a reduction of about 58%. The context window for the hosted version of Clef-flash was also reduced from 64k to 24k.
ModelInput priceOutput price
Clef-flash$0.038 per million tokens (approximately 0.25 yuan at the current exchange rate)$0.12 per million tokens (approximately 0.8 yuan at the current exchange rate)
Clef$0.24 per million tokens (approximately 1.6 yuan at the current exchange rate)$0.72 per million tokens (approximately 4.8 yuan at the current exchange rate)
Clef-omni$0.15 per million tokens (approximately 1 yuan at the current exchange rate)$0.60 per million tokens (approximately 4 yuan at the current exchange rate)