TypeSafe launched Jev because sequential LLMs are “totally useless for computers”
The New Stack Adrian Bridgwater ● Covered by 11 sources
TypeSafe launched Jev, a model that picks decisions instead of writing text. It’s built for software, and TypeSafe says it’s far faster and cheaper than LLM chat.
Based on reporting by The New Stack, Adrian Bridgwater — read the original for the full story.
Summary, retelling and take written by AI under human oversight; images are AI-generated illustrations. How we work · Report an error
TypeSafe came out of stealth last week with $40 million in seed funding led by DCVC and a very pointed thesis: the thing the industry spent years perfecting after ChatGPT may be the wrong tool for computers. Its first model, Jev, is not meant to draft replies or code. It is meant to make decisions inside software.
The company calls Jev a System One model, and that label is doing a lot of work. Developers can send it structured questions and get typed answers back with probabilities attached, so the rest of the application can decide how much trust to place in the result. TypeSafe says it built a new architecture for the model, a new sampler, and a training method called Reinforcement Learning for Calibrated Decisions, or RLCD.
Diogo Almeida, TypeSafe’s co-founder and CEO, used unusually blunt language in the launch. On X, he argued that sequential LLMs are “totally useless for computers” because they generate one token after another, which he says is fine for conversation but slow for software. He also framed Jev as a response to a bigger question: why haven’t superhuman chat models turned into AGI?
TypeSafe’s pitch is speed, cost, and reliability. Almeida’s launch post described Jev as 20-200x faster and 40-400x cheaper, with output tokens free. He also said the model produces decisions with probabilities and confidence rather than words, and claimed that makes it unlike a chatbot and less prone to hallucination.
The early reception seems to have been strong. Vercel said Jev reached more than twice as many paid teams as any previous model launch within 24 hours of appearing on AI Gateway, and that nearly 13% of paid teams were using it by hour 24. Software developer Flavio Copes, meanwhile, described it as a “smart if statement” that sits inside normal code rather than replacing the whole application.
That narrower role is the real story here. Jev is not trying to be another assistant that talks a lot. It is trying to be the little decision engine you plug into a larger system when one judgment matters more than a paragraph.
My take — AI-written commentary, not fact-checked reporting
This is a healthy correction to the “chatbot everywhere” era. Not every problem needs a model that writes a little essay before doing the job. The industry keeps dressing up simple decisions as grand intelligence, and then acts surprised when the output is slow, chatty, and occasionally made of soup.
Read more about this at: The New Stack