From Launch to $7.5 Billion in Three Weeks: TypeSafe AI Closes Roughly $870 Million Led by Andreessen Horowitz

TypeSafe AI, the San Francisco startup behind the Jev decision model, has raised roughly $870 million at a $7.5 billion valuation, according to a Bloomberg report on October 9, with Andreessen Horowitz leading the round and general partner Martin Casado joining the board. Several follow-on reports said Sequoia Capital and a group of angel investors also took part, though Bloomberg's original account confirmed only the a16z lead. The deal came together within weeks of Jev's public release — a striking pace even by this year's standards.

Jev is not built to chat, draft emails or write code. TypeSafe calls its approach "System One Models": the model plugs directly into other software and returns structured decisions — classifications, scores or choices from a fixed set — along with probabilities and confidence levels, at high volume and low latency. The company was founded in 2024 by Diogo Almeida, a former OpenAI researcher whose early work on instruction-following helped lay the groundwork for ChatGPT-style assistants, and it says its team numbers just over 20 people.

The adoption figures circulating with the round are all company-reported and unaudited. TypeSafe says Jev topped one million users within days of its September 15 early-access release, that about a third of the Fortune 500 already use it, and that it now processes trillions of tokens per day. A Wall Street Journal report earlier this month, which this site covered, put the same metrics at roughly one trillion tokens daily and about a quarter of the Fortune 500 — the numbers have grown in the company's own telling as the fundraise progressed.

The timing also marks a step down from the rumor mill. The Information had reported the company was in talks at a valuation above $10 billion; the round closed at $7.5 billion per Bloomberg. At roughly $870 million against $7.5 billion, the implied dilution is about 11.6 percent if the round is all-equity — terms beyond the headline numbers have not been disclosed. The capital is earmarked, per the company, for expanding deployment, hiring and compute.

TypeSafe is the name brand in a category that has multiplied in weeks. OpenAI opened its own Decisions API built on the Luna model; Cloudflare open-sourced Clef, a 27-billion-parameter decision model that claims lower latency than Jev on several benchmarks; Amazon's Strands Labs released the open-source Decider 2B; Databricks shipped an ai_decide function. The pitch across all of them is the same: a large share of machine-to-machine AI work is not generation but judgment — route, approve, flag, score — and it rewards small, fast, calibrated models over general-purpose ones.

The caveats are as consistent as the pitch. None of Jev's performance or customer claims has been verified by an independent benchmark or audit, and the company has not disclosed financials, its website address or the round's full investor list in public reporting. Speed comparisons against frontier models are also not like-for-like: a model that answers from a narrow action space will always beat a general model at that narrow task.

What the round really prices is a structural bet: that the decision layer inside enterprise software is a distinct market from the chat layer, and that whoever owns calibration at scale owns a toll booth on the path to agentic computing. a16z has now put the largest check behind that thesis. Whether it holds up will depend on whether independent benchmarks — and audited revenue — eventually materialize to match the trillion-token narrative.