Meet Cloudflare Clef-omni: A 30B-A3B Multimodal Decision Model for Audio, Video, Image and Text
Cloudflare has launched Clef-omni, a multimodal decision model designed to evaluate content across four modalities - text, images, audio, and video - in a single inference call. The model is built on top of Qwen3-Omni-30B-A3B, a mixture-of-experts architecture that activates roughly 3 billion parameters per forward pass despite having 30 billion total parameters. This design keeps compute costs relatively low while preserving broad capability across modalities.
The "decision model" framing positions Clef-omni as a tool for automated moderation, content classification, and policy enforcement pipelines, rather than a general-purpose generative assistant. By handling all four modalities in one call, developers can avoid stitching together separate specialized models for audio transcription, image classification, and text analysis, which typically adds latency and operational complexity.
Cloudflare is offering Clef-omni at $0.15 per million input tokens, with a faster and cheaper variant called Clef-flash priced at $0.038 per million input tokens. The flash tier likely trades some accuracy or context depth for throughput, giving teams a way to run high-volume screening at lower cost before escalating uncertain cases to the full model. Both models are released under the Apache 2.0 license, meaning the weights can be used commercially and modified without royalty obligations.
The release fits into a broader pattern of infrastructure providers moving AI inference closer to their existing network and developer tooling. Cloudflare's Workers AI platform already hosts a range of third-party models, and adding a proprietary omni model gives the company a differentiated offering in the content safety space. For teams already running workloads on Cloudflare's edge network, integrating Clef-omni means one fewer external API dependency for moderation tasks.
