TypeSafe AI has raised $870 million in a deal that values the whole company at $7.5 billion, just weeks after it released its first public AI model, Jev, in mid-September. Andreessen Horowitz, the venture firm also known as a16z, led the round, with Sequoia Capital and DCVC, an earlier backer, taking part. TypeSafe says a16z partner Martin Casado joins its board. The startup is led by its founder and CEO, Diogo Almeida.
Jev is not a chatbot. Most AI tools people know, such as ChatGPT, write text for people to read. Jev answers the narrow questions a program asks while it runs: which option from a list to pick, how to score something against a set of rules, or whether a statement is true. It returns the answer in a fixed form the program can use directly, along with how confident it is, and TypeSafe says the software can then act on its own when Jev is confident and ask a person when it is not. TypeSafe calls this a System One model and says each question should be a gut-check judgment, the kind an expert could make in a few seconds.
a16z points to price and speed. TypeSafe charges 4.2 cents per million tokens Jev reads, the small chunks of text, often parts of words, that AI models read and write. Its answers are so small that it does not charge for them, a16z says. The firm says Jev costs roughly 1/100 to 1/500 as much as the most capable AI models and is 100 times faster at sorting things into categories, with similar accuracy. Those figures come from TypeSafe and the firm now invested in it, and no one else has shown them yet.
a16z, which led the round and so has a stake in Jev's success, calls Jev the fastest-growing AI model it has ever seen, saying it produced 1 trillion tokens in its first three days. The company and its investor give different figures for big business: TypeSafe says a third of the Fortune 500, the biggest U.S. companies, use Jev, while a16z says 25% have built it in. Neither names the companies or says how much they pay.
The money arrives as Microsoft moves into the same idea. On Oct. 9, Microsoft released Microsoft-Decision-1, its own model for quick yes-or-no, multiple-choice and rating decisions inside software, at the same price of 4.2 cents per million tokens it reads, with its answers free. Microsoft says these decision models are quickly becoming an important new category in AI, and that its model was the most accurate and the fastest in its own comparison of them.
TypeSafe has not said how it will spend the money beyond promising more models built for software and the features big customers have asked for, and it has not shared its revenue. It is not yet clear how Jev and Microsoft's model compare on customers' real work.