Oct 3, 2026 · 8 min · 7 segments
Saturday commute briefing. Cloudflare ships Clef and Clef-flash — open decision models that return yes-or-no and pick-a-lane answers so agents can act without a paragraph; pilot them on the hot path…
Cloudflare trained two models it calls Clef and Clef Flash and put them on Workers' AI.
That's their service for running models close to users on the edge network, and also published the weights under an Apache license, which means you can download and run them yourself.
These are decision models.
A decision model is not a chatbot.
You give it a situation and typed questions.
Is this urgent? Which team should own it? How bad is the impact? And it returns probabilities your code can act on.
Route the ticket, escalate, classify a domain, or hand off to a human.
Cloudflare says its own threat intelligence team already uses Clef to put websites into categories like fashion, e-commerce or phishing in about half the time a general large language model needed for the same job.
The models take images as well as text, and they speak the same API shape as a rival product called Jev, so shops already wired for that pattern can try a swap.
There is a fine-tune path if you want Cloudflare's engineers to specialize it on your labeled decisions.
I'm bringing Tim in on whether Monday is a pilot for the hot path, the fast step that sits in the middle of every agent request or a shrug that classifiers already exist.
This is about decision models from Cloudflare, and what I mean is tools that answer bounded questions with probabilities so an agent can act without writing a paragraph.
That is a real shortlist change for teams that today burn a top-shelf chat model, the expensive general ones on yes or no routing.
On Monday, if agents classify, escalate, or route in your stack, stand up a Clef or Clef Flash pilot on Workers' AI or on your own kit with the open weights and keep a general language model for open-ended reasoning.
Don't use the expensive open model for every fork in the road.
We've had classifiers for years.
Why is this not a rebrand?
Classic classifiers usually need a fresh train when you invent a new category.
These take open-ended questions and still return typed answers you can wire into code.
If your agent only needs a lane and a confidence, that is a different product from a chat model that might invent a third option in prose.
So pilot Clef on the hot path, keep a general model for judgment.
Cloudflare trained two models it calls Clef and Clef Flash and put them on Workers' AI.
That's their service for running models close to users on the edge network, and also published the weights under an Apache license, which means you can download and run them yourself.
These are decision models.
A decision model is not a chatbot.
You give it a situation and typed questions.
Is this urgent? Which team should own it? How bad is the impact? And it returns probabilities your code can act on.
Route the ticket, escalate, classify a domain, or hand off to a human.
Cloudflare says its own threat intelligence team already uses Clef to put websites into categories like fashion, e-commerce or phishing in about half the time a general large language model needed for the same job.
The models take images as well as text, and they speak the same API shape as a rival product called Jev, so shops already wired for that pattern can try a swap.
There is a fine-tune path if you want Cloudflare's engineers to specialize it on your labeled decisions.
I'm bringing Tim in on whether Monday is a pilot for the hot path, the fast step that sits in the middle of every agent request or a shrug that classifiers already exist.
This is about decision models from Cloudflare, and what I mean is tools that answer bounded questions with probabilities so an agent can act without writing a paragraph.
That is a real shortlist change for teams that today burn a top-shelf chat model, the expensive general ones on yes or no routing.
On Monday, if agents classify, escalate, or route in your stack, stand up a Clef or Clef Flash pilot on Workers' AI or on your own kit with the open weights and keep a general language model for open-ended reasoning.
Don't use the expensive open model for every fork in the road.
We've had classifiers for years.
Why is this not a rebrand?
Classic classifiers usually need a fresh train when you invent a new category.
These take open-ended questions and still return typed answers you can wire into code.
If your agent only needs a lane and a confidence, that is a different product from a chat model that might invent a third option in prose.
So pilot Clef on the hot path, keep a general model for judgment.
The rest of this transcript — segmented and speaker-labeled, so you land on the exact moment something was said
Search every transcript — by keyword, by phrase, or by meaning, across every show Radar indexes
Trends — what is surging across podcasts, measured against its own baseline
Alerts — when a name you follow appears in a newly indexed episode
No account is needed to search Radar.
