Cloudflare launches Clef-omni and cuts Clef-flash to $0.038: near-free AI decisions for agent builders
Redacción IvokaOct 11, 9:00 AM 3 min
Cloudflare cut the price of its cheapest decision model by 58% and launched a version that understands audio, video, images and text in a single call.
On October 9, Cloudflare announced three changes to its "Clef" family of decision models. Instead of writing text, these models return scored options so an agent can choose what to do (for example, which tool to use or how to classify a message).
What's new
- Clef-flash dropped from $0.09 to $0.038 per million input tokens. To get there, Cloudflare shrank the hosted version's context window from 64k to 24k tokens; the company says only 0.24% of requests went past that limit.
- Clef-omni is new: it accepts audio, video, images and text in a single call, at $0.15 per million tokens. Cloudflare says it scores a 21-second clip with sound in about 1.5 seconds.
- Clef (the base model) stays at $0.24 and now responds roughly twice as fast after infrastructure changes.
The weights are open, so anyone with bigger needs can run it themselves. Price and speed figures come from Cloudflare, not independent testing.
Why it matters
If you don't code, you won't use this directly. But it signals that the "deciding" part of agents (classifying a customer message, picking a reply, routing a quote) is getting very cheap, and the tools you use may get cheaper over time.
Sources