TypeSafe AI raises $40M seed led by DCVC for its non-chat decision model Jev
The AMW Read
Confirms funding behind an already-observed production-adoption pattern (Vercel Gateway) and introduces a schema-constrained, non-generative decision model as a distinct new-entrant category within foundation models.
TypeSafe AI raises $40M seed led by DCVC for its non-chat decision model Jev
San Francisco-based TypeSafe AI ended two years of stealth on September 15 with the public launch of Jev, the first model in what it calls its "System One Model" line, and has now disclosed a $40 million seed round led by DCVC. Forbes reported, citing a person familiar with the deal, that the round valued the company at $200 million — a figure TypeSafe AI has not confirmed. The company was founded in 2024 by CEO Diogo Almeida, who co-developed OpenAI's RLHF pipeline and worked on InstructGPT, ChatGPT, and GPT-4, alongside co-founders Erik Gafni and Sasha Sheng. Almeida said Jev was trained exclusively on synthetic data through the stealth period.
Jev is not a conversational LLM. It returns typed, schema-constrained outputs with attached probabilities — for example a fire-risk probability on an insurance policy — generated in parallel rather than token-by-token, which TypeSafe AI says cuts response latency to 70-500 milliseconds versus 3 to 329 seconds for frontier chat models, priced at $0.042 per million input tokens with no charge for output tokens. The company's own workflow benchmarks claim up to 193.6x faster and 444.6x cheaper results than incumbent LLMs, though those figures are self-designed and unverified by any third party; TypeSafe AI itself cautions real-world gains vary by workload and region. Its "no hallucination" claim is narrower than it sounds — it guarantees schema-conformant output, not correct answers. Per the AI Market Watch index, TypeSafe AI has now appeared in 2 pipeline-ingested items in the past 90 days against zero in the prior 90, consistent with a coverage cluster forming around Jev's launch rather than a one-off story.
The funding follows adoption rather than preceding it — extending AMW's September 20 coverage of Jev becoming the fastest-adopted launch in Vercel's AI Gateway history. A Vercel engineer said swapping an OpenAI-based command-safety classifier for Jev delivered 5-18x faster results at higher accuracy, and Bryo AI's CTO said Jev was the only model in an email-classification comparison to return probabilities reflecting true confidence, versus Gemini's slightly higher accuracy at 10-20x the cost. For builders running narrow, schema-bound classification or scoring steps inside larger agent or LLM pipelines, Jev is a concrete option worth benchmarking on cost and latency — but neither the speed multiples nor the valuation should be treated as independently confirmed.
