Skip to content
AI.info

The Pulse

Mirendil raises $1 billion for AI research tools

Kleiner Perkins and Clearlake Capital led Mirendil’s Series A funding round. The company is developing models and agents to help scientists and engineers conduct AI research.

Mirendil raises $1 billion for AI research tools

Mirendil has raised $1 billion in a Series A round led by Kleiner Perkins and Clearlake Capital to pursue tools for AI research and development, the company said in a post on November 12.

Battery Ventures, Doerr Capital, Gil & Co, Google, M13, Pear VC and S32 also participated, alongside other new and existing investors. Mirendil is building tools intended to make advanced AI research accessible to scientists and engineers.

Model built with agents

In an October account of its work, Mirendil said one researcher working with agents pretrained its coding model, Mirendil-Base, in 3.5 months using less than $10 million in compute. The company said agents wrote over 99% of the code, ran experiments and diagnosed failures, while the researcher chose what to try and intervened when needed.

The compute figure includes research, small-scale experiments, the final training run, agents’ token spending and data processing, according to Mirendil. The training data came from publicly available code, web pages and datasets; the company said it did not purchase or license additional data or use synthetic data or distillation for the experiment.

Mirendil subsequently post-trained the base model into Fern, which supports a context window of up to 10 million tokens. Fern 0.1 takes text and code as input and has a size of 0.1–0.5 trillion parameters.

Private beta for Lor

Lor, the company’s tool for working with agents, is in private beta by invitation. It is free during the beta and available on macOS, Linux and the web, with Windows, iOS and Android versions listed as coming soon.

The company says Lor agents can write code, run experiments and work for days, including by splitting tasks across many agents. The tool provides compute that agents can request and release, and it can use machines selected by the user.

In one example described by Mirendil, agents built environments from 100 PyTorch pull requests using 102 agents that averaged 8 minutes each. Lor supports Fern as well as models from other providers, including Anthropic, OpenAI, Google and DeepSeek.

Sources

Related stories

Explore

More articles