Mirendil, a startup building artificial-intelligence tools that could vastly accelerate the pace of AI research and development, has raised $1 billion in its latest funding round.
Venture-capital firm Kleiner Perkins and private-equity firm Clearlake Capital led the Series A round, Mirendil said Thursday. The company raised a $200 million seed round less than six months ago, another example of how VCs are rapidly deploying massive amounts of capital into a small group of AI startups.
The company didn't disclose the valuation at which it raised the funding, though people familiar with the matter said the Series A round valued Mirendil at $5 billion.
Battery Ventures, Doerr Capital and Gil & Co. also participated, Mirendil said, alongside others. Bloomberg earlier reported some details of the funding round.
Co-founded by a group that includes two Anthropic alumni who departed the AI giant at the close of 2025, Mirendil is developing AI that can improve and build upon itself to accelerate the pace of AI research. That, in turn, could help democratize the capabilities now reserved largely for frontier AI models. It could also enable small teams of researchers as well as scientists and engineers to develop their own in-house AI models for specialized purposes, co-founders Behnam Neyshabur and Harsh Mehta told The Wall Street Journal earlier this year.
The billion-dollar fundraise comes as private markets show some unease with sky-high valuations of AI companies. OpenAI, for example, raised less than half of its $30 billion target for its latest mammoth funding round, and at a valuation of $1.25 trillion -- down from the $1.4 trillion valuation it had been seeking, the Journal reported. Similarly, Anthropic has tamped down the anticipated valuation of its IPO from as high as $2 trillion to $1.33 trillion.
To be sure, investors are still willing to bet big on promising AI startups.
Mamoon Hamid, a partner at Kleiner Perkins, which also co-led Mirendil's seed round, said in a post on the social-media platform X that Mirendil's team had "exceeded even our highest expectations" over those six months.
Mirendil late last month said that one of its researchers had built its first pretrained model over the course of 31/2 months, using $10 million worth of compute. Pretraining, or the early stages of developing an AI model, has historically required "hundreds" of researchers in frontier labs, the startup said in its announcement.
The company also unveiled Lor, an AI research-and-development platform it said was composed of hundreds of agents writing code and running experiments for days at a time.
"What excites us most isn't just what Mirendil can accomplish as an AI lab," Hamid wrote. "It's what this approach could unlock for everyone else."
He cited drug discovery and robotics as among the fields that could benefit from Mirendil's platform.
Some voices in the AI community have voiced alarm over the idea that self-improving AI could improve at speeds that would effectively make it impossible for humans to control. The Mirendil co-founders have said they believe it is possible to safely oversee the technology.