In the past couple of weeks, one AI has been everywhere: Jev, from a new company called TypeSafe AI. The headlines call it "the AI that can't chat", which is confusing. Why would anyone want that? Here's the short answer, and then the details.
The AI you already know, like ChatGPT or Claude, is a writer. You ask something and it writes an answer, one word after another. Jev doesn't write anything. You give it some information and a multiple-choice question, and it ticks an answer and tells you how sure it is. Think of it less as a writer and more as a very fast, well-read person filling in a form.
TypeSafe calls this a "System One" model, after the idea that people have a fast, gut-feeling way of thinking (System 1) and a slow, careful one (System 2). Chat AIs are the slow, careful kind. Jev is built to be the quick one.
An email arrives: "My order still hasn't turned up and I want my money back." Your software asks Jev three questions at once:
In well under a second, Jev might answer: billing (fairly sure), urgent (very sure), and 7 out of 10. Your software can act on that straight away. A chat AI could answer too, but it would take longer, cost more, and might reply with a friendly paragraph that your software then has to untangle.
ChatGPT runs on OpenAI's GPT models, and Claude is made by Anthropic. Both are writers that can also do quick sorting jobs, so they're the natural thing to compare Jev with. Prices are the ones each company published in September 2026.
Best forFast, repeated choices from a list you wrote
Doesn't write anything. It picks an answer from your options and says how sure it is, in a fraction of a second.
Best forWriting, reasoning and code
Anthropic's AI. Talks with people, explains, writes code and works through hard problems. Can sort emails too, but writes its answer out word by word.
Best forWriting, reasoning and code
OpenAI's AI, running on its GPT models. Does the same kinds of jobs as Claude, and its smallest model is the cheapest writer here.
We took the email example above and ran the numbers for 1,000 emails, assuming about 500 tokens (roughly 375 words) going in for each email and its questions, and a short answer of about 50 tokens coming back from the chat AIs. Model by model:
So Jev is far cheaper than the big chat AIs, but OpenAI's smallest model gets surprisingly close, at about four times the price. Treat these as ballpark figures: each company counts tokens a little differently, and chat AIs that "think" before answering use more tokens than this.
On speed, Jev's 0.07 to 0.5 seconds comes from TypeSafe. In TypeSafe's own tests, the most capable chat models took from about 3 seconds to several minutes on the same jobs. Small chat models like Claude Haiku and GPT-6 Luna are much quicker than that, but still slower than Jev claims to be, because they write their answer out.
They also work well together: an app could use Jev to sort thousands of things quickly, and hand only the ones that need a written reply to Claude or ChatGPT.
Good fits: anything with lots of small, repeated decisions. Sorting support tickets, spotting spam, flagging risky messages, scoring sales leads, checking another AI's answer before it goes out, or making split-second choices in games, robots and simulations.
Not a fit: a chatbot, a writing helper, research, or anything where you want the AI to explain itself.
No. Jev is a paid service that runs on TypeSafe's computers, and it's in early access for now. You can't download it or run it yourself, which is why it doesn't have a page in our directory.
There is a community project called open-jev that rebuilds the idea from scratch, but it's an untrained experiment for learning how this kind of model might work, not something you can use for real jobs. If you want something free and private today, the closest option is to run an open AI model on your own computer with Ollama and ask it to answer in a fixed format. It's slower than Jev, but nothing leaves your machine and there's nothing to pay per answer.
If your app makes the same kind of small decision over and over, it's worth a test. Take a hundred real examples where you already know the right answer, run them through Jev and a small chat AI, and compare how often each gets them right, how long it takes and what it costs. If you're looking for a helpful assistant to talk to, Jev isn't that, and it isn't trying to be.
No. It does a different job. Chat AIs write, explain and hold conversations. Jev only makes quick choices from options you set up in advance. Many apps could use both: a chat AI to talk to people, and Jev for the many small decisions behind the scenes.
It can't invent an answer that isn't on your list, because it only ever picks from the options you give it. But it can still pick the wrong option. That's why it tells you how sure it is, so you can send the unsure cases to a person.
Not of Jev itself. A community project called open-jev rebuilds the idea for learning purposes, but it isn't trained, so it can't do real work. If you need something free and private today, you can run an open model on your own computer with Ollama and ask it for answers in a fixed format. It's slower, but nothing leaves your machine.
TypeSafe AI, a San Francisco start-up led by a former OpenAI researcher. It launched Jev on 15 September 2026 and has raised $40 million.
Get up and running with Kimi, GLM, MiniMax, DeepSeek, gpt-oss, Qwen, Gemma and other models.
