TypeSafe AI releases Jev to automate software tasks
Startup TypeSafe AI has launched Jev, a non-language model that outputs decision probabilities to provide developers with a faster, cheaper, and hallucination-free automation tool.
TypeSafe AI, a startup founded by former OpenAI researcher Diogo Almeida, has released a new transformer-based model called Jev. Unlike traditional large language models, Jev does not generate text. Instead, it outputs calibrated decision probabilities, making it highly optimized for computer-to-computer automation. Because users predefine the potential outputs, the model is incapable of hallucinating. Furthermore, its output tokens are entirely free, while input tokens are billed by the billion rather than the million.
Early adopters report significant performance and cost improvements. Pranit Sharma, a software engineer at Vercel, noted that replacing OpenAI's ChatGPT Luna 5.6 with Jev for a safety command classifier yielded results five to 18 times faster, alongside improved accuracy. Similarly, Bryo AI CTO Nikhil Mudholkar compared Jev to Google's Gemini for email classification. While Gemini was slightly more accurate, it was 10 to 20 times more expensive than Jev, which also provided valuable confidence scores for workflow automation.
Beyond direct classification, developers like Armin Ronacher, the CTO of Earendil, which builds the open-source model harness Pi, suggest using Jev for real-time model routing and monitoring other AI agents to prevent jailbreaks. Almeida describes Jev as a "System One model" focused on intuition rather than complex reasoning. The model is trained entirely on synthetic data using a proprietary technique called reinforcement learning from calibrated decisions. Almeida hopes the low cost of the model will lead to emergent, distributed smart software across the internet.
This is our own summary of reporting by TechCrunch AI
