Cloudflare undercuts Jev as its maker raises $870M

The money went to TypeSafe. The pressure went to its price.

On Oct. 9, TypeSafe AI announced a giant round for Jev. The same day, Cloudflare shipped an open-weight rival that it says is API compatible and, in one tier, cheaper.

A $7.5B round and an open-weight rival on the same day

TypeSafe says Andreessen Horowitz led an $870M Series A at a $7.5B valuation, with Sequoia Capital and existing investor DCVC joining. Martin Casado of a16z joins the board. The company calls Jev its first "System One Model." It is in early access, and TypeSafe publishes no price on the post.

TypeSafe says about one-third of the Fortune 500 are using Jev. That figure is the company's own.

Cloudflare's post launched Clef-omni, a decision model that takes text, images, audio (WAV or MP3) and video (MP4 or WebM). Cloudflare says it is built on Alibaba's Qwen3-Omni-30B-A3B-Instruct, a mixture of experts model, with the speech output parts removed. The weights are on Hugging Face, and the hosted version costs $0.15 per million input tokens on Workers AI.

It also cut Clef-flash:

From $0.09 to $0.038 per million input tokens.

Cloudflare says that now makes Clef-flash cheaper than Jev, and that Clef is "fully Jev-API compatible," so switching means changing the model ID.

Decision models skip the writing

A decision model returns a choice, not paragraphs. Cloudflare says Clef outputs decisions and skips output token generation. That is why the price is quoted on input alone, and why latency is the pitch: Cloudflare claims about 130 ms median for text and about 1.5 seconds for a 21-second video with sound.

Here is the verdict. A $7.5B valuation prices Jev as the default. Cloudflare just made it a line item you can shop.

Be careful with the scoreboard, because both sides keep it. On Cloudflare's own table, Clef-omni beats Jev on BFCL (98.2 to 95.75) and BANKING77 (94.8 to 79.74). Jev leads on When2Call (80.97 to 63.3). On TypeSafe's evals, which Cloudflare also ran, Jev matches or beats Clef-omni on every workflow listed.

So: cheaper and open on one side, stronger on its home turf on the other.

The fine print is the context window

Cloudflare cut the hosted Clef-flash context window from 64k to 24k tokens. It says only 0.24% of requests ran past 24k. The open weights were trained for 256k and are unchanged if you host them yourself.

If you feed long documents into a router, that change matters more than the price.

What to do, if you use a model to route support tickets, pick a tool or classify intent:

1. Pull a few hundred real requests from your logs.
2. Run them through Jev and through Clef-flash or Clef-omni with the same prompt.
3. Compare accuracy, latency and the bill, and check how many of your inputs exceed 24k tokens.

Your traffic is the only benchmark that counts here.

Watch Jev's price

TypeSafe has not published one in its funding post. With an open, compatible rival at $0.038, it will have to.

Watch the price page. Then run the swap test.