Cloudflare Releases a Model for AI Agent Decisions Without a Human
Cloudflare has released the Clef and Clef-flash models for AI agent decisions made without human input. Clef-flash is more than 10 times faster than the rival Jev model.
Recently Cloudflare released their newest models, Clef and Clef-flash. These models do not generate text, but rather select the most probable response out of several choices that you provide (e.g., “urgent” or “not urgent,” “team A” or “team B”). This means that an agent can make autonomous decisions without needing human intervention (e.g., clicking “approve”). As an example, Cloudflare describes that their support tickets are classified based on urgency and assigned to the appropriate team automatically using these models. The primary selling point here is speed — Clef-flash processes inputs in 39 milliseconds compared to Jev by TypeSafe AI, which takes 524 milliseconds. In other words, Clef-flash is over 10 times faster!
In situations where we want to classify or route things, we don’t necessarily need a large, complicated model like the ones that I use in Claude Code. Instead, we can use smaller, faster models like Clef and Clef-flash. The weights (i.e., the trained model files) are available under an Apache-2.0 license on Hugging Face (a website for hosting AI models) so that you can deploy them yourself and replace any overused full LLMs with simpler yes/no models. While it may sound like humans aren’t needed in the loop, keep in mind that these are basic tasks such as routing and classification, not an agent making real decisions on your behalf.
Source: the-decoder.com
Free course
Stop reading about AI — start building with it
The free Claude Code course: your first site, tool or game — no coding. No upsells, no cross-sells — nothing to buy here.
Start free →
Author
Evgenii Arsentev
PhD · Founder & CEO, Dad4Kids
Articles · Latest articles