Overview
- TypeSafe exited stealth with a $40 million seed round led by DCVC and introduced Jev as a purpose‑built model from founder Diogo Almeida, who previously worked on reinforcement learning at OpenAI.
- Jev does not generate text or code and instead emits three fixed primitives — Choice (select from up to 255 options), Score (numeric rating), and Noul (boolean probability) — each returned with a confidence value.
- TypeSafe published pricing that charges $0.042 per million input tokens while making outputs effectively free because Jev does not use autoregressive text generation.
- The company released benchmarks and a demo showing rapid call rates (an example video shows roughly ten calls per second and an estimated $7 hourly cost) and claimed up to roughly 20–200× speed and 40–400× cost advantages versus frontier LLMs, but those figures are company‑provided and not independently verified.
- TypeSafe positions Jev for tasks such as risk scoring, moderation routing, and automated workflow branching, which could cut costs for repeated low‑latency decisions if third‑party tests confirm the firm’s speed, cost, and calibration claims.