TypeSafe Inc., the company behind the Jev artificial intelligence model, said it has closed an $870 million funding round at a $7.5 billion valuation. Andreessen Horowitz led the round, with participation from Sequoia Capital, DCVC and unnamed angel investors. The raise arrives less than a month after TypeSafe launched Jev, and the company said roughly a third of the Fortune 500 has already adopted the model.
Jev is built to solve a problem that arises when enterprise applications use large language models. Such applications send a description of a task to a model, which typically replies in natural language text that the application must then convert into a structured, standardized format before it can be used. Jev skips that step by generating structured output directly, sparing applications the reformatting work.
The model accepts only three request types. An application can ask Jev to answer a question in a yes-or-no equivalent, choose an item from a list, or produce a score. Developers can define what the score measures, so Jev can be set up to rate the severity of cybersecurity alerts or gauge how urgent a support ticket is.
Because Jev's short answers are easy for applications to handle, developers write less data preparation code, which TypeSafe says lets software projects finish faster and lowers the risk of errors. The model also reports a number reflecting its confidence in a response when it picks from a list or generates a score, giving applications a way to limit the effects of hallucinations.
TypeSafe said it trained Jev with a new method it calls reinforcement learning for calibrated decisions, a variant of the widely used reinforcement learning approach, along with a new model architecture. The company claims those technologies let Jev handle requests in under 700 milliseconds, making it up to 200 times faster than some frontier LLMs, and up to 100 times more cost-efficient.
Jev belongs to a planned model series called System One. TypeSafe will spend part of the new funding on expanding that lineup, and it also intends to release enterprise features, which it did not detail, aimed at making the models easier for large organizations to adopt.
More company and startup news from TechManNews.
