TypeSafe AI raises $870M at $7.5B valuation weeks after Jev launch

TypeSafe AI lands $870 million weeks after Jev's debut

TypeSafe AI, the startup behind the non-text AI model Jev, has closed an $870 million round at a $7.5 billion valuation, according to TechCrunch. Andreessen Horowitz led the financing, with Sequoia and existing backer DCVC also taking part. The raise arrives only weeks after Jev launched on September 15 and spread quickly.

A transformer that speaks in probabilities

Jev is built on a transformer architecture, but TypeSafe is explicit that it is not a large language model. Instead of producing text, the model outputs probabilities for different choices — the company brands these results "calibrated decisions" — a design aimed at deciding what a process should do next rather than writing prose or code.

TypeSafe claims Jev runs significantly faster than LLMs and consumes far fewer tokens, which sits at the centre of its pitch: automation workloads do not need a model optimised for human language. Co-founder Diogo Almeida, previously a researcher at OpenAI, told TechCrunch last month that AI has become highly capable at human language over the past four years, but that this skill is not what automation needs, because computers communicate in a different way.

A dev.to post (originally published by Hacks.gr) corroborates the outline: Jev returns probabilities rather than text or code, and the new capital is intended to fund further development of the model.

Enterprise adoption claims

The most striking figure in the story is adoption. TypeSafe says a third of Fortune 500 companies are already using Jev — an unusually fast pace of enterprise uptake for a product that has been publicly available for under a month. Both TechCrunch and the dev.to item attribute this figure to the company itself, and it has not been independently verified.

If the claim holds up, it would indicate that large organisations are actively shopping for automation models that avoid the cost and latency profile of general-purpose language models.

The people behind it

TypeSafe AI was founded in 2024 by three people: Diogo Almeida, the former OpenAI researcher; Sasha Sheng, a former Meta research engineer; and Erik Gafni, an engineer and entrepreneur. DCVC's status as an existing investor points to earlier funding, although the sources do not detail previous rounds or the total capital raised to date.

Why it matters

This is one of the largest recent bets on an AI model that is deliberately not an LLM. A $7.5 billion valuation for a company whose flagship product launched weeks earlier signals that top-tier investors see a credible alternative to text generation in automation: models that emit decisions rather than language.

The stakes are architectural as much as commercial. Much of the current AI stack assumes general-purpose language models are the right substrate for automating work. If TypeSafe's claims on speed and token efficiency prove out at scale — and if the Fortune 500 adoption figure is real — Jev becomes a data point that the industry can split automation workloads away from the LLM layer entirely, with implications for how enterprises budget, deploy and evaluate AI systems going forward.