Mirendil Raises $1 Billion for Long-Running AI Research

Nathan Brooks

Mirendil has raised $1 billion in Series A financing to expand tools that let AI agents code and conduct experiments over extended periods. Kleiner Perkins and Clearlake Capital led the round, Mirendil said. Battery Ventures, Doerr Capital, Gil & Co, Google, M13, Pear VC and S32 participated alongside other new and existing investors.

The backing finances a specific shift in AI tooling: from assistants handling isolated tasks to agents sustaining technical work for days. Mirendil said smaller research teams could use that approach for projects that would otherwise require substantially more personnel. The company also wants to make advanced AI research and development more accessible to scientists and engineers.

The round gives Mirendil room to advance its agent model, broaden access to frontier AI development, evaluate early models and bring Lor to more users through private beta. Mirendil, which describes itself as a small team, is recruiting people to support the effort.

Mirendil’s announcement set out its central ambition: “enabling individuals and small teams to pursue work traditionally associated with larger organizations.” That is the practical proposition behind the financing, according to the company.

A coding model, post-training and a multi-agent environment

Mirendil highlighted three recent developments. Mirendil-Base is its coding model; Fern is the subsequently post-trained model; and Lor is an agent-based research environment now in private beta. According to Mirendil, the products together extend agents from individual assistance into infrastructure for sustained technical work.

The clearest proof point is Mirendil-Base. According to Mirendil, one researcher working with agents pretrained the coding model from scratch in 3.5 months using less than $10 million in computing resources. The result puts a concrete time-and-compute marker behind Mirendil’s claim that small teams can pursue work traditionally associated with larger organizations.

Fern is the subsequently post-trained model. It supports up to 10 million tokens of context, which Mirendil describes as “sufficient to hold days of an agent’s work.” For research teams, the practical change is continuity rather than isolated assistance, which is how Mirendil describes its strategy.

Lor extends the system into a multi-agent environment. Mirendil says hundreds of agents can write code and run experiments for days at a time. The company also uses Lor in its own daily operations, while the environment remains in private beta.

What changes for small research teams?

For scientists and engineers, the change is the ability to delegate sustained research operations to coordinated agents. Mirendil cites one researcher using agents to pretrain a coding model in 3.5 months, while Lor can have hundreds of agents write code and run experiments for days. The company says the approach can reduce the staffing required for research projects.

For Mirendil, the financing provides capital for a development push that remains early. Mirendil says it remains a small team. Lor remains in private beta, and the early models require evaluation as Mirendil hires. Its stated uses for the money map directly to those tasks: advance the system, widen access and strengthen the team.

The strategic bet is that agents become infrastructure for technical work, not merely software for isolated tasks. Mirendil said its products are designed to operate across days and, in Lor’s case, across hundreds of agents.

For Mirendil, the next checkpoint is execution: advance the models, widen private-beta access and add people while keeping the multi-day workflow intact. Those steps will show whether the $1 billion financing can turn Mirendil’s early models and private-beta environment into a broader research platform.