IT Home On October 10, Cloudflare posted an announcement yesterday (October 9) stating that it has launched the open-weight decision model Clef-omni, which supports processing text, images, audio, and video through a single API call.
In terms of positioning, this model is based on the previous Clef series models, with extended support for multi-modal input. The model weights have been made available on Hugging Face.
In terms of multi-modal support, Clef previously supported text, images, and continuous frames extracted from videos (extracting the video into static images over time, and passing these images in order as input to the model). Clef-omni adds audio and full video input, supporting formats such as wav, mp3, mp4, and webm.
Cloudflare says that developers do not need to set up a separate speech-to-text and audio/video splitting process, and can use one model to handle various inputs.
Cloudflare said that Clef-omni is built based on Qwen3-Omni-30B-A3B-Instruct and retains its main understanding ability. The model is designed for structured decision-making tasks and does not generate regular text outputs. In the tests published by the company, the response time for pure text requests is approximately 130 milliseconds, for images around 150 milliseconds; for a 21-second video with sound, scoring takes about 1.5 seconds.
| Baseline Test | Clef-omni | Clef | Clef-flash | Jev |
|---|---|---|---|---|
| BFCL: Exact Match Rate | 98.2 | 98.47 | 98.76 | 95.75 |
| ToolRet：nDCG@10 | 66.6 | 69.19 | 66.43 | 65.28 |
| API-Bank: Accuracy | 92.7 | 91.93 | 93.11 | 88.19 |
| Home appliances: Exact Match Rate | 69.3 | 82.95 | 97.73 | 52.27 |
| When2Call: accuracy rate | 63.3 | 72.37 | 65.58 | 80.97 |
| BANKING77: Macro Average F1 | 94.8 | 94.20 | 90.93 | 79.74 |
| CLINC150+OOS: Average Macro F1 | 97.7 | 97.43 | 66.77 | 89.27 |
| BRIGHT：nDCG@10 | 42.0 | 45.91 | 39.26 | 47.52 |
| Amazon ESCI: Macro Average F1 | 57.8 | 57.48 | 57.39 | 55.21 |
| PhishNChips: Accuracy | 73.2 | 79.60 | 75.05 | 62.55 |
Cloudflare will also reduce the price of Clef-flash’s input Token per million from $0.09 (IT Home note: current exchange rate is approximately $0.6 in RMB) to $0.038 (current exchange rate is approximately $0.25 in RMB), a reduction of about 58%. The context window synchronization for the Clef-flash hosting version will be reduced from 64k to 24k.
| Model | Enter price | Output price |
|---|---|---|
| Clef-flash | $0.038 per million tokens (approx. 0.25 RMB at current exchange rate) | $0.12 per million Tokens (approx. 0.8 RMB at current exchange rate) |
| Clef | $0.24 per million Tokens (equivalent to about 1.6 RMB at current exchange rate) | Cost per million tokens: $0.72 (equivalent to approximately 4.8 yuan at current exchange rate) |
| Clef-omni | $0.15 per million tokens (approximately 1 RMB at current exchange rate) | $0.60 per million tokens (equivalent to about 4 RMB at current exchange rate) |