Pomegra
TypeSafe AI raised an $870M Series A at a $7.5B valuation led by Andreessen Horowitz, 24 days after Jev's launch, to expand its decision-model developer tool.
Key Takeaways
- TypeSafe AI raised $870M at a $7.5B valuation on October 9, 2026, led by Andreessen Horowitz.
- The round follows a $40M seed led by DCVC, estimated at a roughly $200M valuation.
- Jev returns typed decisions instead of text, priced at $0.042 per million input tokens.
Lead
TypeSafe AI, a US startup that emerged from stealth in September 2026, announced an $870M Series A on October 9 at a $7.5B valuation. Andreessen Horowitz led the round, with Sequoia Capital, existing investor DCVC and a group of angel investors participating. The company will spend the money on its Jev model, additional models, infrastructure for what it calls smart software, and enterprise features.
The speed is the story. Jev reached early access on September 15, and the Series A came 24 days later. Martin Casado, a general partner at Andreessen Horowitz, is joining the board.
What Is Jev and What Does It Do?
Jev is a decision model that takes a typed question, evaluates it against a supplied state and returns a structured answer without generating or parsing text. TypeSafe describes it as the first entry in a model class it calls System One, built for fast, structured decisions inside software.
The model exposes three output types: a choice among supplied options, a numeric score against a rubric, and a true-or-false probability. TypeSafe reports response times of 70 to 500 milliseconds and says that is 40 to 200 times faster than frontier language models on comparable tasks. Those figures come from the company and have not been independently verified.
Typical uses include support ticket classification, invoice validation, security alert triage, contract risk scoring and routing between AI agents. Jev displaces custom classifiers, rules engines and general-purpose language models called through function-calling.
How Is Jev Priced?
TypeSafe charges $0.042 per million input tokens and nothing for output tokens. Because answers are short typed values rather than prose, the free output tier removes the cost that dominates most language model bills.
One customer case study cited by the company shows an 88% drop in cost per screening and 50% faster processing at recruiting startup Jack & Jill. TypeSafe also says Jev generated 1 trillion tokens within three days of launch and that a third of the Fortune 500 use it. Andreessen Horowitz has put that figure closer to 25%, and neither number has been audited.
Why Did the Valuation Jump So Far So Fast?
The valuation rose roughly 37-fold in under a month because investors are pricing adoption velocity, not revenue history. Sequoia has said TypeSafe hit a $100M annualized revenue run rate within seven days of launch. A run rate extrapolated from one week of usage-priced API traffic is a thin base, and it can reflect trials and one-off batch jobs as much as recurring contracts.
At $7.5B, the round implies about 75 times that run rate. That multiple assumes usage compounds and that customers keep sending volume once pilots end. The seed round, led by DCVC at an estimated valuation near $200M, was priced before any of this traffic existed.
The size of the raise also matters. $870M is large for a Series A and suggests that training compute, not headcount, is the main use of capital. Andreessen Horowitz is betting that a narrow model class can hold margin against general-purpose labs that can cut prices.
How Does Competition Look for Typed-Decision Models?
Competition is already arriving. Cloudflare released Clef on October 1, an open-weight model under the Apache 2.0 license that the company reports at 38.8 milliseconds latency. An open alternative at that speed pressures the premium TypeSafe can charge, even at $0.042 per million tokens.
Frontier labs could also fold structured-output modes into existing models and compete on price. TypeSafe's defense is architectural: a model that cannot emit free text cannot hallucinate prose or break a parser. That claim is credible for format errors. It does not address whether the decision itself is correct, which is the harder audit problem for enterprise buyers.
Strategic Context
Founder Diogo Almeida previously worked at OpenAI on instruction-following research. His pitch targets software that makes thousands of small judgments per second, where latency and token cost matter more than conversational fluency. That market exists today inside rules engines and human review queues, but its size as a model-driven category is unproven.
Outlook
TypeSafe now holds $870M, a board seat for Andreessen Horowitz and a valuation that demands sustained growth. The next tests are whether early usage converts into committed enterprise contracts, whether Jev's speed advantage survives open-weight rivals like Clef, and whether TypeSafe ships a second System One model on the schedule its funding implies. Independent verification of the Fortune 500 adoption claim would also settle the gap between the company's figure and its lead investor's.
