TypeSafe AI, founded in 2024 by Diogo Almeida, Erik Gafni, and Sasha Sheng, has launched its Jev model. The startup secured $40 million in seed funding led by DCVC. Almeida previously spent four years at OpenAI, contributing to projects including ChatGPT and GPT-4.
Jev is designed for structured decision-making tasks like email classification and transaction flagging. The model processes requests in 70 to 500 milliseconds, with a median response time of 0.44 to 0.48 seconds. It is trained exclusively on synthetic data using a method called Reinforcement Learning for Calibrated Decisions.
Pricing for the model is set at $0.042 per million input tokens, with output tokens provided for free. Benchmarks conducted at Vercel indicate that Jev is 5 to 18 times faster than certain OpenAI models and 10 to 20 times cheaper than Google's Gemini for specific classification tasks.
Following its launch on September 15, the model experienced high demand that overloaded its API. The system is closed-source and focuses on deterministic outputs rather than creative generation. Developers have noted the reliability of its confidence scores as a key feature for automation pipelines.
