TypeSafe AI

TypeSafe AI is an artificial intelligence lab in San Francisco, founded in 2024 and run by chief executive Diogo Almeida alongside co-founders Erik Gafni and Sasha Sheng. Almeida was a researcher at OpenAI, where he worked on reinforcement learning from human feedback, InstructGPT, ChatGPT and GPT-4. The company came out of stealth on September 15, 2026 with a 40 million dollar seed round led by DCVC. On October 9, 2026 it raised 870 million dollars at a 7.5 billion dollar valuation, in a Series A led by Andreessen Horowitz with Sequoia Capital and DCVC. Martin Casado joined its board, and TypeSafe says a third of the Fortune 500 already use Jev.

Its thesis is that the industry has built almost all of its models for conversation, while most of the work those models are actually asked to do happens inside software with no person reading the answer. A program classifying a support ticket, checking whether an agent should be allowed to run a tool, or grading a draft against a rubric does not need prose. It needs a value. Asking a chat model for that value means paying for generated text, waiting for it, parsing it back into a usable type, and handling the cases where the format comes back wrong.

TypeSafe calls its answer a System One model, a name borrowed from the fast, automatic kind of human thinking. Its first one, Jev, takes a block of state and a set of typed questions and returns structured answers directly: a choice from a named set, a score on an ordered scale, or the probability that a yes-or-no question is true. Each answer carries a calibrated confidence, several questions can share one block of state in a single request, and nothing is generated as text. The company trained it with a method it calls reinforcement learning for calibrated decisions, and produces every output in parallel rather than one token after another, which is where its reported response times of 70 to 500 milliseconds come from. Input is priced at 0.042 dollars per million tokens and output is not billed.

The company is at an early stage and should be read that way. It has one model, a small team and, since October, a very large Series A, but still no independent evaluation of the speed and cost figures it publishes. Its own interface remains a waitlisted early access program with no announced date for general availability, and the practical way to reach Jev is through a gateway: Vercel AI Gateway listed it on September 16, 2026 without a waitlist, and Cloudflare carries it in Workers AI. The idea was a genuine departure, and it was copied fast. Within three weeks an open model from Convai Innovations had arrived, and on October 1, 2026 Cloudflare released its own Jev-compatible Clef models under an open license while Amazon Web Services released Strands Decider. Almeida says he sees no real competition yet. Whether a lab this size can hold its lead against open imitators, and against the same capability arriving as a feature of a frontier chat model, is the open question.