Mirendil Raises $1B Series A Led by Kleiner Perkins and Clearlake

What happened

The round was led by Kleiner Perkins and Clearlake Capital, with Battery Ventures, Doerr Capital, Gil & Co, Google, M13, Pear VC and S32 among the participants; the company disclosed no valuation. At its June 24 launch, Mirendil disclosed a $200 million seed round led by Andreessen Horowitz and Kleiner Perkins at a $1 billion valuation, with Nvidia among the follow-on investors.

Mirendil is led by CEO Behnam Neyshabur and CTO Harsh Mehta, who helped automate parts of Anthropic's internal research; the founding team of 20 researchers and engineers comes from Anthropic, xAI, Google DeepMind and OpenAI. The company says its aim is to let one person take on work that used to need a team and a team take on work that used to need an institution, by handing the hands-on work of frontier AI research to agents.

An Oct 21 report said one researcher working with agents pretrained the Mirendil-Base coding model from scratch in 3.5 months on under $10 million of compute, then post-trained it into Fern 0.1, which holds up to 10 million tokens of context, enough for days of an agent's work. The Lor platform, introduced on Oct 28, is in private beta, with hundreds of agents writing code and running experiments on it; Mirendil says the company runs on Lor every day.

Key facts

1. Round: $1 billion Series A led by Kleiner Perkins and Clearlake Capital; no valuation disclosed
2. Investors: Battery Ventures, Doerr Capital, Gil & Co, Google, M13, Pear VC and S32, among others
3. Seed: A $200 million seed round disclosed at its June 24 launch, led by Andreessen Horowitz and Kleiner Perkins at a $1 billion valuation
4. Team: CEO Behnam Neyshabur and CTO Harsh Mehta; a founding team of 20 from Anthropic, xAI, Google DeepMind and OpenAI
5. Pretraining: One researcher working with agents pretrained the Mirendil-Base coding model from scratch in 3.5 months on under $10 million of compute; the model sits in the 100-billion-to-500-billion-parameter range
6. Products: The post-trained Fern 0.1 model holds up to 10 million tokens of context; the Lor agent research platform is in private beta