Mirendil Raises $1 Billion to Sell the Idea That One Researcher With Agents Can Pretrain a Model
Mirendil said on November 12 that it raised a $1 billion Series A led by Kleiner Perkins and Clearlake Capital, with Battery Ventures, Doerr Capital, Gil & Co, Google, M13, Pear VC and S32 among the investors. The company, which emerged in June, says one researcher working with agents pretrained its Mirendil-Base coding model from scratch in 3.5 months on less than $10 million of compute, and it sells Lor, a private-beta platform that runs hundreds of agents on research work for days. Mirendil did not disclose a valuation, and its training claims have not been independently checked.
Mirendil's $1 billion is a bet on a claim about labour: that agents doing the hands-on work mean "the size of your team stops limiting what you can build." The company announced on November 12 that it had raised a $1 billion Series A led by Kleiner Perkins and Clearlake Capital, less than five months after it emerged with a $200 million seed round, according to Unite.AI.
The round#
Battery Ventures, Doerr Capital, Gil & Co, Google, M13, Pear VC and S32 also took part, among other new and existing investors, Mirendil said. It did not disclose a valuation or how much each investor put in. At its June launch, Mirendil described a founding team of 20 researchers and engineers from Anthropic, xAI, Google DeepMind and OpenAI, and a seed round led by Andreessen Horowitz and Kleiner Perkins, Unite.AI reported. It also has a multi-year Google Cloud partnership announced in August.
What it says it has built#
Mirendil says one researcher working with agents pretrained Mirendil-Base, a coding model, from scratch in 3.5 months with less than $10 million in compute. Unite.AI, summarising Mirendil's October 21 report, says the company claims agents wrote more than 99% of the code, and that the figure covers experiments, the final run, agent token spending and data processing. Mirendil then post-trained the base model into Fern, which it says holds up to 10 million tokens of context.
Its product is Lor, now in private beta: "hundreds of agents writing code and running experiments for days at a time," the company says, adding that it runs on Lor itself. The pitch sits alongside other recent claims of agent-built software, such as AWS's emulator that it says no human wrote or reviewed.
Limits#
- The one-researcher training result, cost figure and Fern's context length are Mirendil's claims; we have seen no independent evaluation of Mirendil-Base or Fern.
- No valuation was disclosed. A jobs newsletter put it at $5 billion; we could not confirm that figure and do not use it.
- Unite.AI's account of the seed round, founding team and Google Cloud deal summarises earlier Mirendil announcements we did not reopen.