TypeSafe raises $870M at a $7.5B valuation, 24 days after launching Jev — venture’s biggest bet yet on AI that decides instead of writes

TypeSafe AI has raised an $870 million Series A at a $7.5 billion valuation, led by Andreessen Horowitz — just 24 days after launching Jev, the transformer that skips text entirely and returns calibrated decisions. It’s the largest venture bet so far on a post-LLM category: machine-native models built for automation, not conversation.

TypeSafe AI has raised an $870 million Series A at a $7.5 billion valuation, led by Andreessen Horowitz — just 24 days after launching Jev, the transformer that skips text entirely and returns calibrated decisions. It’s the largest venture bet so far on a post-LLM category: machine-native models built for automation, not conversation.

The round was announced on Friday, October 9. Alongside Andreessen Horowitz, Sequoia Capital, existing investor DCVC, and a group of angel investors participated, and a16z’s Martin Casado joins TypeSafe’s board, according to the company’s announcement and its law firm’s client highlight, which dates the announcement to October 9. Wilson Sonsini advised on the transaction.

24 days from launch to $7.5B

Jev launched on September 15 and went viral almost instantly. In its own announcement, TypeSafe says a third of the Fortune 500 are now using it and that it has already saved customers “millions of dollars in production.” The round follows a $40 million seed led by DCVC that was announced on launch day, per Bloomberg’s reporting. The Information had earlier reported TypeSafe was discussing a raise of $1 billion or more at valuations of at least $10 billion; the closed Series A landed at $870 million and $7.5 billion.

The model that decides

Jev is built on a transformer architecture, but it is not a large language model. It doesn’t output text; it produces probabilities — what the company calls “calibrated decisions.” TypeSafe says it trained Jev with “Reinforcement Learning for Calibrated Decisions,” or RLCD, and describes the approach as a “System One Model.”

In practice, that means three request types: a yes-or-no answer, a pick from a list, or a score — for example, rating the severity of a cybersecurity alert or quantifying the urgency of a support ticket. The structured output removes the step where applications must condense an LLM’s natural-language text into a usable format before acting on it.

Why investors are paying up

The thesis belongs to co-founder Diogo Almeida, a former OpenAI researcher who worked on InstructGPT: “We have been super good at human language for four years, but it’s not useful for automation because computers speak a different language,” he told TechCrunch last month. He co-founded TypeSafe in 2024 with former Meta research engineer Sasha Sheng and engineer-entrepreneur Erik Gafni.

The bet is economic as much as technical. TypeSafe’s pitch is that Jev works significantly faster and uses far fewer tokens than LLMs — positioning the approach for automating tasks rather than generating text or code. If machine-readable decisions become the primitive for agent workflows, the inference economics of the whole automation stack change.

What the round signals

A $7.5 billion valuation for a 24-day-old model says investors now treat “decision models” as a venture category of their own — one where structured outputs with calibrated probabilities replace parsed LLM prose as the unit of work. The open questions are the ones every breakout model faces: whether the calibration holds up in real customer workflows, and whether the company’s enterprise claims — a third of the Fortune 500, millions saved in production — survive contact with audited numbers. For now, the market has priced the category.