Cloudflare's open Clef decision models return a routing label in about 39ms. Ticket triage just got cheaper to build

Decision models went from one product to a competitive category in a few weeks. That is good news for anyone routing leads or tickets.

What happened

Cloudflare released Clef and Clef-flash, two decision models hosted on Workers AI and open sourced on Hugging Face under Apache 2.0. Instead of writing text, a decision model returns typed answers with probabilities. Cloudflare's example: pass in a support message and ask whether it is urgent and which team should handle it. Your code then routes the ticket, escalates it, or defers to a human.

Cloudflare reports median latency of about 39 milliseconds for Clef-flash and about 209 milliseconds for Clef, compared with just over 524 milliseconds for TypeSafe AI's Jev, according to The Decoder. Clef accepts images, which Cloudflare says Jev does not yet, and has a 64k context window versus Jev's 32k. The models are Jev API compatible, so switching is meant to be simple. Cloudflare is also launching a reinforcement learning service so customers can fine-tune Clef on their own tasks, initially handled with its engineers.

Cloudflare's own claim is bold: "a human does not necessarily need to be in the loop for agentic decisions anymore." These benchmarks are Cloudflare's own.

My take

The useful part for operators is not the speed race. It is that the decision step in a workflow is becoming a commodity you can swap.

Think of the lead or ticket routers you already run. Most use a big chat model to produce one label, then parse its text. A decision model gives you the label and a probability, which is exactly what a router needs.

How I would use it:

• Act automatically only above a confidence threshold you choose
• Send anything below it to a human queue, with the probabilities attached
• Log every decision so you can measure accuracy against what people actually did

On the "no human in the loop" line, I would keep a person on low confidence and high stakes calls, like refunds or account changes. Probabilities tell you when to ask.

Ticket triage, lead routing, reporting, data entry, content pipelines, agent workflows: I design and build automations that do these jobs end to end. See the range at romielwillautomate.dev.