Clef-flash

Clef-flash is the smaller, latency-optimised variant of Cloudflare’s Clef decision models, released on 1 October 2026. Like Clef, it returns typed, bounded answers with a probability for each option instead of free text, so that agents can make routine classification decisions without a human in the loop, and it accepts text or images with a 64,000-token context. Cloudflare says it is built on a frozen Qwen 3.5 9B backbone. The weights are on Hugging Face under the Apache 2.0 licence, and the model runs on Cloudflare’s Workers AI; the announcement gives no price. It suits high-volume decisions where response time matters more than the last points of accuracy, with the larger Clef available for harder cases.

Not yet independently verified: Every detail here comes from Cloudflare’s own announcement; no independent report was found. Cloudflare says Clef leads a third-party decision benchmark, which we have not read, and the post gives no Workers AI price. We will update this when it can be confirmed, and remove this note.