A tour of TypeSafe AI's new decision-making model, Jev, and the eight things people already built with it in its first week: email sorting, ad blocking, instant video clipping, and more.
Jev is not a text-generation model. It is a decision engine that scores structured yes-or-no and classification questions at a fraction of a cent, and that speed and cost advantage, not raw intelligence, is what makes it useful.
Jev is a new AI model from TypeSafe AI, built by ex-OpenAI researcher Diogo Almeida, designed to make structured decisions fast and cheap rather than to generate text. It costs about 4.2 cents per million input tokens, with unlimited free output tokens, a price its maker claims is up to 238 times cheaper on input than Claude. The video tours eight things people built with it in its first week: detecting AI-written 'slop' on websites, an ad and cookie-banner blocker, an email inbox that sorts by importance instead of arrival time, a fuzzy find-in-page tool, a webpage that assembles itself from a component library in real time, a tool that clips long videos into topic-based cuts in under two seconds, a color-palette generator, and an emoji matcher. The throughline: Jev doesn't replace chat models, it's a cheap decision layer for structured problems chat models are too slow or expensive to solve at volume.
Sign in and you get 23 free chat messages on us — ask for the hook, quote a framework, find the exact transcript moment, generate a markdown action plan. Bring your own key when you want unlimited.
Create a free account →
Cold open comparing Jev's moment to the early days of ChatGPT, promises an explanation of what it is plus a run of things people have already built with it.
Explains the TypeSafe AI blog post announcing Jev, walks through a structured JSON decision prompt (is a hot dog a sandwich, resolved 73 percent true), states when Jev works best (defined decisions, clear context) versus when it isn't a fit (text generation, unstructured questions, personal advice), then demonstrates the speed-over-intelligence tradeoff with a timed blitz chess match against Claude Fable 5.1 and covers pricing.
Shows an AI slop detector built on Jev scoring anthropic.com at 26 percent AI slop, breaking down the specific signals it checked.
Demos Unclutter, a Chrome extension by @thekitze that uses Jev to auto-clean ads, cookie banners, upsells, and pop-up dialogs off any page it loads, BYOK and open source.
Zapier added Jev as a native app in its automation builder. Demos wiring a Google Calendar new-event trigger into a TypeSafe Jev action that decides whether to accept or decline the invite by set criteria, usable across Zapier's 9,000-plus integrations.
Shows a tweet from @jnnnthnn: an inbox concept that sorts by importance instead of chronological order, built with Jev, sorting 100 emails by priority in under half a second.
Demos a fuzzy find-in-page Chrome extension by @Saboo_Shubham_ that matches meaning instead of the exact typed string, tested against Apple's terms and conditions page.
Shows a generative-UI demo (json-render plus Jev, by @ctatedev) assembling a working dashboard interface from a pre-built component library in milliseconds, picking layouts and fields based on what the user needs.
Demos Clip Fast, a tool by @BurhanUsman that scans a 90-plus minute video and cuts a topic-specific clip in under two seconds for about two cents.
Shows a tweet from @mattdesl demoing Jev generating color palettes from plain text prompts like tomato, neon tokyo, browser, and Mario characters.
Shows a tweet from @heystefan_ demoing Jev matching typed phrases to the emojis that represent them, then Berman closes with a pointer to a deeper follow-up video on Jev.
Jev's entire value comes from being fast and cheap enough to make structured decisions at a volume and price no general chat model can touch, not from being smarter than one.
“Jev is everywhere and it is the most exciting AI that I have seen in a long time.”
“Jev is really good when you can describe the type of decision you want to make, when you can provide really clear context and structured questions.”
“If you put Jev up against Fable in a game of chess with a time limit, Jev will most likely win just because Fable takes so long to process each step and Jev can make a decision nearly instantly.”
“Jev is 4.2 cents per million input tokens and free for output tokens, unlimited output tokens, absolutely free.”
See every word as it's spoken — crank it to 2× and still catch all of it. The same dual-channel trick behind Amazon's Kindle + Audible.
Berman compares the moment to the earliest days of ChatGPT: something powerful and raw that nobody has fully figured out yet. Before the demos start, he draws a hard line around what this new model is, and is not, actually for.
Instead of an open-ended question, define the decision as a JSON object: state the instruction, then give explicit true and false criteria the model should check against. Jev returns a confidence score against that rubric rather than a free-text answer.
“It's Zapier. I've been working with them for a while. They actually added Jev into the Zap platform.”
Mid-video sponsor block tied directly into the video's own subject: instead of a generic ad read, Berman demos Jev running as a native Zapier action, so the sponsor segment doubles as another use-case demo.
00:00
00:11
00:1900:30
00:35
00:43
00:5101:02
01:07
01:15
01:23
01:31
01:39
01:47
01:57
02:03
02:12
02:19
02:27
02:35
02:43
02:51
02:59
03:0703:1503:23
03:29
03:39
03:47
03:55
04:0304:14
04:19
04:27
04:35
04:43
04:50
05:00
05:07
05:17
05:22
05:30
05:4205:46
05:54
06:02
06:10
06:1906:27
06:31
06:42
06:5006:58
07:03
07:14
07:22
07:30
07:38
07:4507:54
08:03
08:10
08:15
08:3008:34
08:42
08:47
08:58
09:0709:14
09:22
09:30
09:38
09:4609:52
10:02
10:10
10:16
10:26
10:34Add Modern Creator as a preferred source and Google shows you more of our breakdowns in Search, Top Stories, and AI Overviews. It only changes what you see, and you can undo it in your Google settings anytime.
12:31ChatGPT co-inventor Diogo Almeida ships a model that's 200x faster and effectively free, but Matthew Berman's own chess test shows it losing on skill, only surviving the clock.
September 18th
26:49A 27-minute systems playbook for turning AI coding tools into a self-managing development flywheel.
June 18th
33:44How one MacBook running Claude Opus 4.6 replaced a CRM, a security firm, a content team, and a personal chef -- with the exact prompts to copy every piece.
February 17th
18:57The mystery model that took over OpenRouter turns out to be a Chinese open-weights release that nearly matches frontier intelligence for a few cents a task.
August 29th
08:54A 'dot' release plays out like a full generational leap: two five-to-seven-day unsupervised coding runs, a sponsor benchmark, and a live pricing and capability standoff against a rawer, higher-ceiling rival model.
July 9th
16:11A 16-minute walkthrough of 7 copy-paste AI coding agent loops — each one runs autonomously until a defined goal is met.
June 19th