Eight hundred seventy million for a model that writes nothing

TypeSafe AI, the developer of an AI model called Jev that does not write text, has raised $870m at a $7.5bn valuation, the company said on Friday. The round was led by Andreessen Horowitz, with Sequoia Capital, existing investor DCVC and a group of angel investors taking part, according to the company's announcement. Martin Casado, an Andreessen Horowitz general partner, is joining the board, the company said.

The model launched in early access on 15 September, so the round came 24 days later, by our count. Jev is built on a transformer architecture, as large language models are, but it is not one. It outputs probabilities, which the company calls "calibrated decisions," TechCrunch said. According to TypeSafe's documentation, it evaluates typed questions against a supplied state and returns typed values and probabilities that code can branch on. It offers three question types: Choice picks one option from a list, Score rates the state against a rubric and Noul returns a probability from 0 to 1 that the answer to a yes-or-no question is yes.

The pitch is that most of what software asks of AI is not conversation. "We have been super good at human language for four years, but it's not useful for automation because computers speak a different language," Diogo Almeida, TypeSafe's co-founder, told TechCrunch last month. Almeida previously worked at OpenAI. TypeSafe was co-founded in 2024 with Sasha Sheng, a former Meta research engineer, and Erik Gafni, TechCrunch said.

The company's numbers are its own. In its launch post it says Jev answers in 70 to 500 milliseconds, and it publishes a price of $0.042 per million input tokens, with output tokens free. The 193.6-times-faster and 444.6-times-cheaper figures on its homepage come from its workflow evaluations, and the company says it expects them to be at the high end of real-world gains. It also notes that its own model capabilities team built those evaluations and that some bias could exist.

One customer case study shows what the claim looks like in use. Jack & Jill, a talent marketplace, replaced Gemini 3.1 Flash Lite for all calls in one stage of its candidate-matching pipeline within 10 days of first testing, according to TypeSafe's case study. Screening cost fell from $0.755 to $0.092 per 1,000 candidates, which by our arithmetic is a cut of about 88 per cent. Median screening time fell from 20.3 seconds to 10.3 seconds. Ranking quality, measured as area under the curve, was 0.933 for Jev against 0.924 for Gemini, and the case study said the difference in how many of the candidates hiring managers later asked to meet each model kept on its shortlist was not statistically significant.

The adoption claim is where the two accounts part. TypeSafe said in its announcement that a third of the Fortune 500 are using Jev. Andreessen Horowitz's own post says 25 per cent of Fortune 500 enterprises have integrated it. The firm also said Jev reached 1 trillion tokens generated within three days of launch, and called it "the biggest narrative violation we've seen this year, in the best possible way." Neither figure was independently verified in the accounts we read.

The firm's case for the deal is about what AI does to software, not about competing with chatbots. Jev is "the first of a new class of System One models," Andreessen Horowitz wrote, "which are built to be deeply integrated with code (rather than talked to by people) that returns instinctive decisions in the native language of software." The "System One" name draws on Daniel Kahneman's distinction between fast and slow thinking, and Jev is named after the economist William Stanley Jevons, the company says.

The valuation moved fast. TypeSafe came out of stealth on 15 September with $40m of seed funding led by DCVC, and Forbes reported at the time that the round valued the company at $200m. By our arithmetic, $7.5bn is 37.5 times that figure.

The cost question runs through our recent coverage of AI economics. Agent task costs can vary up to 30-fold between runs, McKinsey's survey found, and product teams are already redrawing the line between what runs always and what runs on demand. TypeSafe's wager, and now Andreessen Horowitz's, is that the next large market for models is the one where nobody reads the answer.