signal/raiseTrack rounds
Series AAI · San Francisco

Mirendil raises $1bn Series A led by Kleiner Perkins and Clearlake for AI research agents

The San Francisco lab, started by former Anthropic researchers, raised five times its June seed round and did not disclose a valuation.

s/rFunding deskNorth America · 2 min read

Mirendil, a San Francisco startup building AI systems that carry out AI research, raised $1bn in a Series A round, it announced on 12 November 2026.

Kleiner Perkins and Clearlake Capital led the round. Battery Ventures, Doerr Capital, Gil & Co, Google, M13, Pear VC and S32 also invested, along with other new and existing backers the company did not name.

Mirendil did not disclose a valuation. TechTimes, citing a Bloomberg report from September, said the company had been in talks to raise up to $1bn at about $5bn, but found no source confirming the final terms.

“We're still a small team with a singular focus.”Mirendil (company statement)

What Mirendil is building

The company says its aim is to put frontier AI research within reach of many more scientists and engineers. It has described three products.

Mirendil-Base is a coding model that the company says it pretrained from scratch in three and a half months for less than $10m in compute. Fern is a model post-trained from it that, according to Mirendil, handles up to 10 million tokens of context. Lor, in private beta, lets users run hundreds of agents that write code and run experiments over several days; Mirendil says it uses Lor for its own work every day.

TechTimes reported that the company says AI agents wrote more than 99% of the code for its pretraining work, which one researcher directed. It said Mirendil-Base has between 100bn and 500bn parameters, that larger models were still in training in October, and that agents on Lor can request, set up and release computing nodes on their own. The publication noted that these performance and cost figures come from Mirendil's own tests and have not been checked by outsiders.

The team and its funding so far

According to TechTimes, Mirendil was founded by former Anthropic researchers, including Behnam Neyshabur and Harsh Mehta, and launched in June with a team of 20 researchers and engineers drawn from Anthropic, xAI, Google DeepMind and OpenAI. The company still describes itself as a small team and did not give a headcount.

TechTimes reported that Mirendil raised a $200m seed round at a $1bn valuation in late June, about 141 days before the Series A, and that Nvidia invested after that round. Google, a participant in the new round, also supplies the company with computing capacity. TechTimes said Mirendil's chief executive told TechCrunch that the Google Cloud agreement was worth more than $100m, although Google's own announcement gave no figure.

A crowded week for AI rounds

The round was the largest of the week in Crunchbase News's tally of the ten biggest funding rounds, ahead of a $750m Series D for Mercor and a $500m financing for AI infrastructure company Torus.

Mirendil's release did not set out how the money would be spent beyond saying it would help the company push further toward making AI research more direct and accessible.

Company website: mirendil.com