Cloudflare has released Clef-omni, a 30B multimodal mixture-of-experts model, on Hugging Face under an Apache 2.0 license. Unlike traditional LLMs, Clef-omni is designed for decision-making tasks, accepting text, images, audio, or video inputs along with typed questions to return probabilities for predefined answers. This model is positioned as an alternative to TypeSafe's Jev, aiming for production use cases with a focus on efficient decision scoring rather than text generation. While strong in intent classification, Clef-omni shows weaknesses in complex reasoning and RAG hallucination detection compared to its predecessor, Clef. AI
IMPACT This specialized model could accelerate adoption of structured decision-making in AI applications, though its reasoning and RAG capabilities require further validation.
RANK_REASON Cloudflare released a new multimodal model, Clef-omni, with a system card. [lever_c_demoted from frontier_release: ic=1 ai=1.0]
- Apache Software License 2.0
- BANKING77
- Clef-flash
- Clef-omni
- CLINC150
- Cloudflare
- GPQA
- Hugging Face
- MMLU-Pro
- Qwen3-Omni
- SGLang
- Workers AI
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →