/
← Voltar ao feed
low

Cloudflare releases Clef-omni multimodal decision model, cuts Clef-flash price and speeds up Clef

1 fonte(s) · 1 atualização(ões) · publicado 09/10/2026 15:27 · coletado há 2 dias

Cloudflare announced Clef-omni, an open-weight decision model that accepts text, image, audio (wav/mp3) and video (mp4/webm) input in a single API call. It is built on a Qwen3-Omni-30B-A3B-Instruct mixture-of-experts backbone with LoRA adapters. Cloudflare also cut the price of Clef-flash to $0.038 per M input tokens, at the cost of a smaller hosted context window (24k instead of 64k). It sped up the hosted Clef by moving serving to SGLang. This is product news with no vulnerability or attack. The article notes internal security-adjacent uses: spam detection, phishing moderation in EmDash plugin libraries, PII scanning for DLP, and malicious domain detection by the threat intelligence team.

Detalhes técnicos

  • Clef-omni runs a prefill pass over the full payload and scores all modalities and schema options at once, without generating output tokens. It uses two-stage attention routing and a built-in lexical grammar for schema-constrained scoring.
  • Reported latency: about 130 ms median for text, about 150 ms for images, a few hundred ms for audio, and about 1.5 s for a 21-second video clip with sound.
  • Pricing per M input tokens: Clef-flash $0.038 (was $0.09), Clef $0.24 (unchanged), Clef-omni $0.15.
  • Clef-flash weights on Hugging Face are unchanged and support a 256k context window for self-hosting. Cloudflare says only 0.24% of requests exceed 24k tokens.
  • Cloudflare submitted a pull request to SGLang (PR #42721) to support Clef, to be released in SGLang 0.5.22.

Linha do tempo

  1. blog.cloudflare.com ↗ · pub. 09/10/2026 15:27 · coletado 19/10 18:00 primeiro registro

    Cloudflare releases Clef-omni multimodal decision model, cuts Clef-flash price and speeds up Clef

    ler a análise desta fonte

    Cloudflare announced Clef-omni, an open-weight decision model that accepts text, image, audio (wav/mp3) and video (mp4/webm) input in a single API call. It is built on a Qwen3-Omni-30B-A3B-Instruct mixture-of-experts backbone with LoRA adapters. Cloudflare also cut the price of Clef-flash to $0.038 per M input tokens, at the cost of a smaller hosted context window (24k instead of 64k). It sped up the hosted Clef by moving serving to SGLang. This is product news with no vulnerability or attack. The article notes internal security-adjacent uses: spam detection, phishing moderation in EmDash plugin libraries, PII scanning for DLP, and malicious domain detection by the threat intelligence team.

    Technical Details

    • Clef-omni runs a prefill pass over the full payload and scores all modalities and schema options at once, without generating output tokens. It uses two-stage attention routing and a built-in lexical grammar for schema-constrained scoring.
    • Reported latency: about 130 ms median for text, about 150 ms for images, a few hundred ms for audio, and about 1.5 s for a 21-second video clip with sound.
    • Pricing per M input tokens: Clef-flash $0.038 (was $0.09), Clef $0.24 (unchanged), Clef-omni $0.15.
    • Clef-flash weights on Hugging Face are unchanged and support a 256k context window for self-hosting. Cloudflare says only 0.24% of requests exceed 24k tokens.
    • Cloudflare submitted a pull request to SGLang (PR #42721) to support Clef, to be released in SGLang 0.5.22.