Cloudflare releases Clef, open decision models that answer with probabilities

Wednesday, 7 October 2026By Sridhar Mukkandi1 min read

On 1 October, Cloudflare released two open-weight decision models, Clef and Clef-flash, under the Apache 2.0 licence. Clef is built on a 27B Qwen model and Clef-flash on a 9B one. Both read text and images with a 64k-token context window, run on Workers AI, and have weights on Hugging Face.

Cloudflare defines a decision model as one that "makes classifications to help agents decide how to act, based on certain probabilities". Instead of writing text token by token, the models return typed answers with probabilities that code can use to route a ticket, trigger an escalation or defer to a person. Across 43 evaluations, Cloudflare reports a median latency of 209.3 ms for Clef and 38.8 ms for Clef-flash, against 524.1 ms for Jev.

So what

The probability is what makes a decision model useful: code can act on a confident answer and hand a borderline one to a person. Check how well those probabilities match reality on your own data before you pick the cut-off.

SourceCloudflare blogblog.cloudflare.com/clef-decision-models/ Go deeper · BookDecide, Don't Generate

Drafted with AI from the original source, then checked and edited by Sridhar Mukkandi. Spotted a mistake? Write to hello@tarkika.com and we'll correct it here.

More from The Signal

Tools AWS open-sources Strands Decider, a 2B model that picks an answer and says how sure it isA small model that runs on a laptop and reports its own confidence is a cheap first gate for agent steps such as routing, tool choice and guardrails. The team's own guidance is the part to copy: below 0.9, confirm the answer or ask a person. Strands Labs on GitHub Models Reflection announces Beam, a 501B open-weight model it says needs 3 to 4 times less computeIf the efficiency claim holds, an open model near the top of the open field at a third of the compute would cut the cost of serving reasoning-heavy agents. Wait for the weights and run your own tests before planning around it. Reflection AI Careers Anthropic commits $100 million to train 10,000 engineers to deploy ClaudeThe scarce skill is no longer calling a model; it is getting one working inside a real organisation. Anthropic says it wants engineers with a record of building with LLMs and of helping others adopt AI, so shipped, adopted projects count for more than certificates. AnthropicAll stories