This availability expands a broader update to Cloudflare's AI suite, which includes a 27B parameter decision model and a native multimodal pipeline in Clef-Omni for processing audio, video, and images. The company recently doubled inference speeds for the family and cut the price of the Clef-Flash model to $0.038 per million input tokens. Internal teams already use these models to identify personally identifiable information and detect malicious domains.
Tuesday, October 20, 2026
Story
Cloudflare Integrates Clef AI Into SGLang for Sub-40 ms Latency
AI8 days ago
Developers can now access the Clef family of decision models via the /v1/systemone API after a technical integration with the SGLang framework. Cloudflare contributed upstream performance optimizations that enable sub-40 ms decision latency for its Clef-Flash model on a single H200 GPU and allow as many as 64 questions to be answered in one prefill. The company is currently preparing to add support for its multimodal Clef-Omni model to the API.
Looking for more context or analysis?
Tilt chat provides reliable, tailored insights, alerts, and automations.