A couple of weeks ago I wrote about how much I liked Jev. It's a model that doesn't chat. It picks an answer from the options you give it and tells you honestly how confident it is. For analytics that was perfect: cheap, fast decisions for routing, classification and picking the next step. Most marketing decisions come down to choosing one of four buttons anyway. I even built it into my own workflows, where it checks my writing.
TypeSafe, the company behind Jev, raised $40 million for it. That was 16 days ago.
In those 16 days, OpenAI rolled out its own Decisions API at DevDay. Then Cloudflare released Clef: open decision models under Apache 2.0, right on Hugging Face. There are two of them: Clef, built on Qwen 3.8-27B, and Clef-flash, built on Qwen 3.5-9B.
And they're compatible with Jev's API. Change one URL in your code and everything keeps working.
The numbers from Cloudflare's blog:
- latency: Clef 209 ms, Clef-flash 39 ms, Jev 524 ms;
- on Jev's own benchmark, Clef wins 8 tasks out of 10;
- 64k context vs 32k;
- the training method has the same name as TypeSafe's.
In short, decision models went from startup to commodity in two weeks.
If you build agents or automate analytics, that's great news. Routing, picking the next step, escalation: all the things we currently burn a big model and whole seconds on can now run on a small open model in 40 milliseconds, with a probability attached.
