- TypeSafe launched a new AI model called Jev that picks answers instead of writing sentences.
- Each answer comes back in well under a second.
- Jev costs far less than typical chat AI models, and its output is free.
- It struggles with counting, math, dates and multi-step instructions.
TypeSafe AI, started by former OpenAI researcher Diogo Almeida, has launched a new kind of AI model called Jev. Jev looks at a piece of text and picks one answer from a list. Apps can use it to make quick choices, like which team gets an email.
Say a customer emails a company's support team. They've spent three days trying to connect Stripe, the service that takes their payments, and it keeps failing. They're losing sales. Jev reads that message, picks technical support as the right team, flags it as urgent and rates how frustrated the customer sounds.
TypeSafe says each answer takes between 0.07 and 0.5 seconds. That's 40 to 200 times faster than typical top AI chat models, which TypeSafe says take between 3 and 329 seconds to give a full response.
AI models charge based on tokens, the small chunks of text they read and write instead of full words. TypeSafe says Jev costs about 4 cents for every million tokens it reads. That's far less than the 20 cents to $10 that regular AI chat models charge for the same amount. Jev's own output is free.
TypeSafe says Jev can't hallucinate. That means it can't give an answer that isn't on the list you set up. That's because it only picks from a list of answers instead of writing free text. Normal chat AI can still write something your software doesn't expect, even when it's very smart.
Jev has real limits. TypeSafe's own documentation says the model struggles with counting things, doing math, comparing dates, and reading instructions that need several steps of interpretation.
Jev is available now in early access. TypeSafe says it's letting developers sign up on its site and bringing people off its waitlist as fast as it can.
