Cloudflare releases Clef-omni, a 30B-A3B multimodal decision model

Cloudflare launched Clef-omni, an open-weight multimodal decision model that scores text, images, audio and video in a single call. The 30B-A3B MoE built on Qwen3-Omni is available on Workers AI at $0.15 per million tokens under Apache-2.0.

• 30B total, 3B active parameters, 64,000-token context, Apache-2.0 license
• Priced at $0.15 per million tokens on Workers AI; Clef-flash cut to $0.038
• Median latency ~130 ms for text, ~150 ms for images, ~1.5 s for 21-second video
• Leads intent classification with 97.7 macro-F1 on CLINC150+OOS