Mirendil, a startup founded by former Google DeepMind and Anthropic researchers, has raised $1 billion in a Series A round led by Kleiner Perkins and Clearlake Capital. Google was among the other investors. Its first product, Lor, automates parts of AI model training by assigning work to hundreds of AI agents, the company says.
Mirendil says one researcher used the software to build Mirendil-Base, a coding model, for $10 million over about 3.5 months, with the agents writing more than 99% of its code. The company says Lor can also fix errors during evaluation; it has not provided independent validation of these claims.
Why it matters
For AI labs, Lor could reduce the specialist work involved in training models, but Mirendil’s account of its capabilities and results has not been independently validated.
