ChatGPT Pioneer's New AI Is 100x Faster — And Can't Make Things Up
It skips chatting entirely to quietly power the apps you already use.
Diogo Almeida spent years at OpenAI helping build the methods that made ChatGPT possible. Then he decided chat wasn't the future of AI. After two years working in secret, his new company, TypeSafe AI, just revealed its first product: a model called Jev that doesn't talk at all.
Instead of writing sentences, Jev makes fast, structured decisions that software can use directly — think sorting incoming support tickets, flagging suspicious payments, or routing emails to the right folder. Because its possible answers are defined in advance, the model literally cannot make things up (a problem known as hallucination). It also tells you how confident it is with every answer, so software knows when to trust it.
The numbers are striking. Jev is 40 to 200 times faster than today's leading AI models, responding in a fraction of a second rather than seconds. It's also dramatically cheaper: output costs nothing, and inputs run about $0.042 per million words' worth of data — a fraction of what rivals charge.
So what's the catch? Jev can't write, chat, or explain itself. It only works for narrow decisions someone has already set up — so it won't replace your chatbot, and it's only available to developers in early access. Still, for businesses drowning in repetitive decisions, that trade-off might be exactly the point.
- TypeSafe AI, founded by a ChatGPT researcher, launched Jev — an AI that makes decisions instead of writing text.
- It's up to 200x faster and far cheaper than today's AI, with output that costs nothing to generate.
- Because its answers must fit pre-set categories, it can't hallucinate — but it also can't chat or explain.
Why It Matters
Cheaper, faster, safer AI decisions could quietly make everyday apps — banking, email, support — smarter and more reliable.