Andreessen Horowitz has led an $870M round for TypeSafe at a $7.5B valuation, less than a month after the startup shipped its Jev model. Sequoia Capital, DCVC and unnamed angels joined the financing.
The speed is the story. Jev launched on September 15 and the company now says roughly a third of the Fortune 500 are already running it in production. Very few enterprise software products clear that bar in three weeks.
What Jev actually does
Jev is built on a transformer architecture, but it is not a large language model. It never writes prose. Instead it returns probabilities – what TypeSafe calls calibrated decisions. Applications can ask for a yes-or-no answer, a selection from a list, or a score whose meaning the developer defines.
The practical effect is that downstream software stops needing a translation layer. When a conventional LLM answers, an application has to parse free text into structured fields before it can act. Jev returns the structured value directly, so developers write less glue code and ship faster.
The model also emits a confidence number with every answer, letting applications throttle or escalate when certainty drops. That is a direct hedge against hallucination, the failure mode enterprises care about most.
Speed as the wedge
TypeSafe says Jev responds in under 700 milliseconds, up to 200 times faster than some frontier models, at a fraction of the cost. The company credits a training method it calls reinforcement learning for calibrated decisions, paired with a custom architecture.
Founded in 2024 by former OpenAI researcher Diogo Almeida, ex-Meta engineer Sasha Sheng and entrepreneur Erik Gafni, TypeSafe is arguing that automation, not conversation, is where the money sits. The Fortune 500 adoption rate suggests buyers agree.
