Mirendil, a research lab founded by former Anthropic researchers, has raised $1 billion in a Series A round led by Kleiner Perkins and Clearlake Capital, with Google among the participants. The company says one researcher, working with AI agents, pretrained a coding model from scratch in three and a half months for less than $10 million in compute. The money will fund its push to put frontier AI research within reach of small teams and single scientists.
What the round buys
The round is unusually large for a Series A, and the investor list shows how seriously big funds are taking the idea. Kleiner Perkins led both this round and the company’s $200 million seed round in June, joined this time by Clearlake Capital as co-lead. Battery Ventures, Doerr Capital, Gil and Co, M13, Pear VC and S32 also took part.
Mirendil did not disclose the valuation, but the Wall Street Journal reported that people familiar with the matter put it at $5 billion. That would be five times the roughly $1 billion valuation attached to the June seed round, which Andreessen Horowitz co-led and Nvidia later joined. The company also signed a multi-year Google Cloud partnership in August.
The experiment behind the raise
The financing follows the first results Mirendil published. In an October 21 report, the company said a single researcher directing AI agents pretrained Mirendil-Base, a coding model in the 100-billion to 500-billion parameter range, from scratch in three and a half months. The company says agents wrote more than 99 percent of the code behind the training run.
Mirendil then post-trained the model into Fern, its first model, which holds up to 10 million tokens of context, enough for days of an agent’s work. Its Lor platform, now in private beta, runs hundreds of agents writing code and running experiments for days at a time. Mirendil says it runs on Lor every day.
The Google connection runs deeper than the investor list. Mirendil signed a multi-year Google Cloud partnership in August to power its training and inference, and Google joined the Series A as a participant. For a lab whose pitch is doing more with less compute, the backing of the cloud it trains on is both a practical asset and a signal.

Our Read
The bet here is on a workflow, not a model. Mirendil is not asking investors to believe in a benchmark score; it is asking them to believe that agents can now do the slow, hands-on work of research itself, and that a lab of about twenty people can therefore move at the pace of an institution.
That is a bigger claim than a new model, and a harder one to verify. None of the headline numbers has been independently checked: there is no outside evaluation of Mirendil-Base or Fern, no public benchmark suite behind the competitive coding model description, and the larger models were still training when the report came out.
Mirendil is not the only new lab raising at this scale. Thinking Machines Lab released its Inkling 2 Small model this month, and the best-funded new labs are now racing each other to prove that smaller teams can do frontier work. The difference is the claim: Mirendil is selling the process, not the model.
Even so, the market is voting with real money. Britain’s Fractile raised $720 million this week to take on Nvidia’s chip crown, and Anthropic’s IPO book is already several times oversubscribed ahead of pricing on November 18. Mirendil’s raise belongs to the same moment: investors are funding labs and their tooling at a pace the industry has not seen before.
Why It Matters
If agents can pretrain models, the cost structure of frontier research changes. A team of twenty no longer needs the budget of a giant to attempt work that once required one, and that lowers the barrier for new labs to enter the field.
The flip side is concentration: the same agents that let small teams build models also let the best-funded labs move faster than everyone else. Money is now the binding constraint, and a $1 billion Series A buys Mirendil the compute and the hiring to test its claim at scale.
Background
Mirendil was founded in late 2025 by Behnam Neyshabur, its chief executive, and Harsh Mehta, its chief technology officer, both formerly of Anthropic. Its founding team of about twenty researchers and engineers came from Anthropic, xAI, Google DeepMind and OpenAI. The lab emerged from stealth on June 24 with its $200 million seed round.
Source: Mirendil’s announcement and the Wall Street Journal’s report.


0 Comments