Diogo Almeida's TypeSafe launches Jev: 20–200x faster decision model

TypeSafe's launch stands out partly because of who is behind it. Diogo Almeida, who co-invented ChatGPT, spent two years in stealth building a new training method (RLCD) and a new type of frontier model. Jev is not a general language model; it is trained to decide, classify, route, and score rather than generate text, and TypeSafe claims it is 20–200x faster and 40–400x cheaper, with output tokens free.

The company frames Jev as a complement to slower “System Two” LLMs, not a replacement. Letting go of strings and chat is the point: parallel sampling, “no hallucination,” and calibration are the trade-offs it offers. Community reactions centered on the same likely use case — replacing LLMs as structured classifiers, judges, or routing policies in production systems where autoregressive generation is unnecessary overhead.

The launch landed on a day with strong announcements from Gemini 3.8 Live and Periodic Labs, yet TypeSafe's launch sat atop Hacker News all day and drew 4.21M views on the announcement tweet. The caveat from the community is that Jev is closer to a constrained or diffusion-like decision model, so the right mental model is less a GPT replacement and more a cheap, calibrated inference engine for structured choices. Whether it earns that role depends on how many production workflows can actually constrain their outputs to predefined formats, and on whether the speed and cost claims hold up outside the launch materials.