TypeSafe Is Out of Stealth

The official out-of-stealth announcement pointing developers at the site for early access to Jev, with a quote from Diogo Almeida's launch thread. The caveat notes it carries no technical detail.

The docs and the models page: this X post is an announcement, while the models entry carries pricing, aliases, and rate limits for jev-1.13.0.

Diogo Almeida, who says he co-invented ChatGPT, posted the thread: two years in stealth on RLCD, a new way to train models, and a new type of frontier AI model released that day as Jev. The visible bullets claim 20-200x faster and 40-400x on a second axis, cut off by X's Show more.

Through the site. The post says to join the frontier and get access on the website, with the link in the TypeSafe profile. Nothing runs from the post itself, so you still need a console.typesafe.ai account and an API key before any of the models or API entries apply.

Sep 15, 2026 at 7:33 PM from the @typesafeai account. The capture records 717.8K views, 213 replies, 258 reposts, 4K likes and 1.2K bookmarks on the announcement, with the founder's thread quoted inside it.

One reply asks whether Jev could work as a cheap, fast semantic layer over Hugging Face telemetry, estimating probabilities like P(legitimate) and P(credential abuse). Another says the timing undercuts the big AI players. Josh Larsen posted a video of getting Jev to play the Chrome dino runner.

No, and it is worth being plain about it. What we captured is the announcement text plus a truncated bullet list: no API shape, no pricing, no benchmark table and no model list. Anything more specific has to come from the docs or the models page.