$7.5 Billion for a Model That Refuses to Talk
TypeSafe AI, the startup behind Jev — an AI model that returns typed decisions instead of written text — has raised about $870 million at a $7.5 billion valuation in a financing led by Andreessen Horowitz, Bloomberg reported Friday. Sequoia Capital and existing investor DCVC participated, and a16z general partner Martin Casado is joining the board. The deal came together just weeks after Jev’s launch video went viral, which TypeSafe co-founder Diogo Almeida said triggered a surge of interest from chief executives, developers, and investors.
The numbers are staggering because they are so fresh. TypeSafe emerged from stealth on Sept. 15 with a $40 million seed round led by DCVC, at roughly a $200 million valuation. Within ten days, The Information reported investors discussing valuations around $10 billion; the round closed 24 days after launch at $7.5 billion, per a launch case study. The San Francisco company, founded in 2024 by Almeida — who co-developed reinforcement learning from human feedback at OpenAI and worked on InstructGPT, ChatGPT, and GPT-4 — with Erik Gafni and Sasha Sheng, says Jev passed one million users within days of launch and is now used by about a third of Fortune 500 companies. The company did not identify any of those customers.
Jev’s pitch is simple and, on its face, correct: most software does not want an essay from a model. It wants a verdict — a label, a score, a yes-or-no with a probability attached. Because Jev’s output shape is fixed ahead of time, it cannot produce an answer outside that schema, and sampling in parallel rather than token-by-token makes it respond far faster. TypeSafe calls these “System One” models, trained with reinforcement learning for calibrated decisions, or RLCD.
The Awkward Timeline: Everyone Else Just Shipped the Same Idea
Here is where the story gets uncomfortable for the new unicorn. The category Jev defined in mid-September has already been colonized by the giants — this week:
On Oct. 6, OpenAI put its Decisions API into public beta: a dedicated endpoint that takes text or images and returns typed answers — a probability, a choice from fixed options, or a score against ordered levels — instead of prose, priced at $0.10 per million input tokens with nothing charged for output. According to Unite.AI, the endpoint runs on GPT-6 Luna, and OpenAI claims it produces answers about 10x faster than its Responses API.
On Oct. 9 — the same day Bloomberg’s report landed — Microsoft launched Microsoft-Decision-1 in Foundry: a decision-scoring model post-trained from Qwen3.5-9B that takes fixed options and returns a calibrated probability for each, at $0.042 per million input tokens with output free — matching Jev’s own input price point, according to The Decoder. Microsoft’s chart claims 83.5% accuracy across 36 benchmarks with 85-millisecond median latency — Microsoft’s chart, Microsoft’s numbers; rivals dispute how the comparisons were measured.
Why this matters: the moat is the question
“Jev, the startup that kicked off the decision model trend in mid-September, has likely had a rough few weeks,” The Decoder observed. “The idea caught on fast, but the tech was quickly adapted and surpassed using open small language models.” There is even an open-source typed-decision judge that ranks number one of 102 models on the System One Mosaic benchmark, with its code and weights public.
So what did a16z actually buy for $870 million? Possibly this: decision models are real infrastructure for the agentic era, and the winner’s prize is not selling API calls — it is becoming the default decision layer inside enterprise software. But note the honest caveats: the Fortune-500 and one-million-user claims come from the company itself, and every benchmark quoted is vendor-supplied. Twenty-four days after a $200 million seed valuation, the category’s creator is now worth $7.5 billion — in a week when Microsoft and OpenAI shipped rivals. That is either a masterclass in category creation or the priciest front-running of commoditization venture capital has attempted in years.
Meanwhile in Tech News:
Also this week: OpenAI’s $20 billion revenue gap, explained — the company now selling typed verdicts has its own numbers to close.
And: GPT-6’s intelligent UI: what ChatGPT actually becomes next — Luna, the model behind the Decisions API, in the consumer spotlight.
