Cloudflare releases Clef and Clef-flash decision models

On October 2, 2026, Cloudflare released Clef and Clef-flash, a pair of open-weight decision models at 27B and 9B parameters. Instead of generating text, the models return typed probabilities, a design meant to let AI agents make structured decisions without an LLM writing prose first. Both are Jev-API compatible and accept image input, and they run on Workers AI with median latencies of 209.3 ms and 38.8 ms respectively.

The following day, The Decoder added detail: both models are built on Qwen and released under Apache 2.0, and they are aimed at letting AI agents make structured decisions without producing text. According to that report, Clef-flash returns classifications in about 39 ms, making it more than ten times faster than TypeSafe AI's Jev, and the release is framed as a challenge to TypeSafe AI's Jev decision model. The report's headline also states Cloudflare's position that the new Clef model means humans no longer need to be in the loop for agent decisions.

So far the story stops at launch: there is no word in either report on availability beyond Workers AI, pricing, or adoption. The two accounts agree on the model sizes, the open weights, the Apache 2.0 license, the Qwen base, Jev-API compatibility and image input. On speed, the numbers appear as 38.8 ms as a median latency on Workers AI and roughly 39 ms for returning classifications; the reports do not say whether those two figures describe exactly the same measurement.