AI That Makes Decisions but Can't Chat: Meet Jev
Jev is an AI that makes decisions instead of writing text, built by a former OpenAI researcher. Learn how it works and what independent tests found.
The 30-second summary
- Jev is an AI model that doesn't chat or write. It picks an answer from options set in advance and tells you how confident it is.
- It was built by TypeSafe AI, co-founded by former OpenAI researcher Diogo Almeida, and spread very fast after its September 15, 2026 launch.
- The biggest speed claims (40 to 200 times faster) come from the company's own tests; an independent test found a much smaller gap.
Jev is an AI that makes decisions, not conversation. It doesn't chat, write emails, create images or write code. You give it a situation and a list of options, and it returns one decision plus how confident it is. It was built by a company led by Diogo Almeida, who spent about 4 years at OpenAI and helped build ChatGPT.
Here is how Jev works, why it is faster than chatbots, how quickly developers picked it up, and what we still don't know.
How does it work?
Chatbots like ChatGPT write their answers word by word, which takes time and costs money. Jev works differently. It takes in a situation, such as a customer complaint, along with a list of options set in advance. Then it returns a decision in one of three forms: it picks an option (Choice), gives a rating (Score), or answers yes or no with a probability between 0 and 1. Every answer comes with a confidence level, meaning how sure it is.
Because it only picks from the list, it can't return an answer outside it. The company says a decision arrives in 70 to 500 milliseconds, because Jev doesn't write word by word but produces everything in one go. TypeSafe calls this kind of model System One, after the fast, intuitive thinking Daniel Kahneman described in his book Thinking, Fast and Slow.
Jev is built to work inside software, not to talk to people. Vercel lists uses like choosing an AI agent's next tool, deciding whether a workflow should continue, retry, ask the user or stop, scoring urgency or risk before an action, and checking other models' answers.
How do we know?
Diogo Almeida is a former researcher at Google Brain and OpenAI. He spent about 4 years at OpenAI working on RLHF (training models with human feedback), InstructGPT and ChatGPT. In 2024 he co-founded TypeSafe AI with Erik Gafni and Sasha Sheng. On September 15, 2026, the company came out of stealth with $40 million in seed funding led by DCVC and released Jev in early access. The launch video got about 40 million views on X, and according to Bloomberg, investors approached the company with offers valuing it at more than $10 billion.
Adoption and speed were measured in different ways. On Vercel's platform, nearly 13% of paid teams were using Jev within 24 hours, twice the GPT-5.6 family, making it the fastest-adopted model in the platform's history. An independent benchmark by AY Automate on 791 decisions with known right answers found Jev 2 to 3.6 times faster, with accuracy close to small models. In an open Pong test by Ably Labs, Jev made 47 decisions in the first 12 seconds versus 2 or 3 for other models, though all models answered correctly 95 to 100% of the time, and they ran at default settings.
And also
The name comes from economist William Stanley Jevons and his paradox: the cheaper something gets, the more of it people use. The listed price is $0.042 per million input tokens, and output is free.
But keep in mind
- The big speed and cost numbers (40 to 200 times faster, and at best 193.6 times faster and 444.6 times cheaper) come from TypeSafe's own evaluations. An independent test found a much smaller gap: 2 to 3.6 times faster.
- In the independent test, Jev was about as accurate as small models, not large frontier ones. On one task sorting requests into 77 categories, the frontier model was about 5 points ahead.
- When people say Jev 'doesn't hallucinate', it means it never returns an answer outside the allowed options, not that its decisions are never wrong.
- Jev is only in early access. There are no public benchmark scores, and the model itself has not been released to the public.
FAQ
What is Jev AI?
Jev is an AI model from TypeSafe AI that doesn't chat or write text. It picks an answer from predefined options, gives it a score, and estimates the probability that the answer is true.
Who built Jev?
Diogo Almeida, a former OpenAI researcher who spent about 4 years there working on RLHF, InstructGPT, ChatGPT and GPT-4, and also worked at Google Brain. He founded TypeSafe AI in 2024 with Erik Gafni and Sasha Sheng.
What are System One AI models?
It is TypeSafe AI's name for models like Jev that return a quick decision instead of long text. The name comes from Daniel Kahneman's idea of fast, intuitive thinking, which he called System 1.
Is Jev really faster than GPT and Claude?
TypeSafe says 40 to 200 times faster, but those are its own tests. An independent benchmark of 791 decisions found it 2 to 3.6 times faster at the median and 4.7 to 7.5 times cheaper than the two cheapest small models, with accuracy similar to small models.
Sources
- 1Jev, an AI Model That Can't Chat, Takes On Bigger RivalsBloomberg (via Yahoo Finance) · 2026
- 2Introducing System One Models & JevTypeSafe AI (official blog) · 2026
- 3
- 4Jev vs GPT and Claude: Independent BenchmarkAY Automate · 2026
- 5TypeSafe AI Emerges From Stealth With $40M in Funding (press release)Business Wire (via Yahoo Finance) · 2026
- 6jev-pong: Jev vs LLMs at Pong (recorded stats)Ably Labs (GitHub) · 2026
- 7Jev (AI model)Wikipedia · 2026
- 8Shut up and calculate: Jev's new AI primitives for codersThe Register · 2026
These are the original sources; we only explain them. Read them yourself.
Want the source?
Comment this word under any of our Instagram posts and we’ll DM you the link to the source.
Instagram postWas this article useful?
