Hey there!
Summary of this email:
- 🗞️ News of the week: Jev, the model that does not generate text, but makes decisions
- Café con Codely recap (watch it here)
- The joke of the week
🗞️ News of the week: Jev, the model that does not generate text, but makes decisions
On Tuesday TypeSafe introduced Jev, a model that does not generate text.
Behind it is one of the creators of ChatGPT, Diogo Almeida.
Instead, it makes decisions. You give it the state of your program (a JSON, a ticket, a request) and you define the possible outputs in advance. Jev returns one of them with its probability.
They call it a "System One Model", after Kahneman's System 1: the fast intuition that does not reason out loud. An LLM would be System 2, the one that thinks step by step and tells you about it.
This change has three consequences:
- 🚀 Speed: it answers in 70 to 500 milliseconds. Since it does not generate token by token, it produces all the outputs in a single parallel pass. TypeSafe says it is 40 to 200 times faster than a frontier LLM on the same tasks.
- 💸 Price: 0.042 dollars per million input tokens and output is free. GPT-5.6 Terra costs 2 dollars for input and 12 for output.
- 🎯 No hallucinations: the output is always one of the ones you defined. It is impossible for it to return a value your code does not expect.
To prove it they made it play Doom with the game state as text: about 10 decisions per second for 7 dollars an hour. In the demo on their website, Jev answers a query in 0.114 seconds and GPT-5.6 Terra takes 8.566 for the same one.
Here comes the nuance. "No hallucinations" does not mean "no mistakes". Jev does not make up a legal citation because it cannot write legal citations, but it can pick the wrong option.
It is also useless for coding. It supports up to 255 options per decision, and the speed comparisons were made with workflows built by their own team.
We are convinced by the idea. Especially for things like computer use.
It is one more piece of the infrastructure for agents, not a replacement for Fable or Astra. The future we see is an LLM that reasons and delegates the small decisions to models like Jev.
It is in early access with a waitlist at typesafe.ai.
Tomorrow we will be live at 9 CEST discussing this story and many more on Café con Codely. On our YouTube, Twitch and X.
☕ Café con Codely recap
This is the top 3 of headlines we discussed in the latest Café:
- Navier-Stokes: OpenAI takes the Millennium Prize medal after it learned that two mathematicians were ahead. Buckmaster and Alpöge had spent a year in private Codex sessions, and OpenAI offered them to sign without the Anthropic co-author or give up the credit.
- Amazon manifesto: Clare Liguori publishes ten Frontier Engineering principles and the first one is that you are the architect, not the typist. The job is no longer to write code, it is to write the intent and verify the output.
- Spotify: cuts Claude Code tokens by 90% with two hooks and Gemini 2.5 Flash. A hook blocks reads over 350 lines and redirects them to a cheap model that returns only a list of bullet points.
You have the full Café recap with links to each story at the moment we discussed it.
Tomorrow we will be live at 9 CEST discussing this week's news on Café con Codely. On our YouTube, Twitch and X. 🙌
And since you've made it to this part of the newsletter, here's the joke of the week, which I know you were waiting for:
Why do programmers use mechanical keyboards? So their code is strongly typed! 😂 😂 😂
Cheers!