A billion-dollar Series A for model development
Mirendil has raised $1 billion in a Series A round to accelerate its work on advanced AI model development, one of the largest early-stage financings disclosed in the sector this year . The round was led by Kleiner Perkins and Clearlake Capital, with participation from Battery Ventures, Doerr Capital, Gil & Co, Google, M13, Pear VC and S32, among other new and existing investors .
The announcement is important not only because of the size of the check, but because of what Mirendil says it is trying to fund: a system for making frontier AI research less dependent on very large human teams . In the company’s framing, the objective is to put AI R&D tools within reach of scientists and engineers who cannot build a full-scale model lab from scratch .
MarketScreener, citing S&P Capital IQ, reported that Mirendil received $1 billion in financing on November 12, 2026, co-led by returning investor Kleiner Perkins and new investor Clearlake Capital Group . That report also says the transaction included convertible preferred stock, a useful detail because Mirendil’s own announcement did not spell out full financing terms .
What Mirendil says the money will accelerate
The company links the round to three closely related pieces of its model-development stack: Mirendil-Base, Fern and Lor . Mirendil says one researcher working with AI agents pretrained Mirendil-Base, described as a competitive coding model, from scratch in 3.5 months with less than $10 million in compute . It then post-trained that base model into Fern, which the company says can handle up to 10 million tokens of context, enough for extended agent work sessions .
Lor is the product layer around that research approach . Mirendil says Lor is in private beta and coordinates hundreds of agents that write code and run experiments for days at a time . EconoTimes reported that Lor access remains invitation-only and that Mirendil lists Fern access as “coming soon,” meaning the headline financing is not yet matched by broad product availability .
SiliconANGLE described Lor as a platform that automates many tasks involved in model training runs by splitting work across hundreds of AI agents . The same report says Mirendil claims Lor can help with the trial-and-error loop of model development, including creating variations, evaluating them and fixing errors when they appear . That is the core commercial promise behind the raise: if Mirendil can compress experimental cycles, it can reduce the time between research idea and usable model.
Why investors are paying for speed
Frontier model development increasingly looks like a balance-sheet contest. Talent, compute and training data are all scarce, and the cost of repeated experiments can punish teams that cannot finance long research cycles. Mirendil’s pitch is that agent-driven workflows can change that equation by letting smaller teams do work that previously required much larger organizations .
SiliconANGLE reported that Mirendil was launched by former Google DeepMind and Anthropic researchers, including CEO Behnam Neyshabur and CTO Harsh Mehta . That background matters because investors are not just funding a product roadmap; they are funding a team with experience in pretraining and internal research automation at major AI labs .
The investor list also shows how strategic the category has become. Google appears as a participant in the Series A, while the round is led by venture and private-equity capital rather than by a single model-platform incumbent . MarketScreener’s account identifies Alphabet Inc. among the participating investors, aligning the transaction with the broader pattern of large technology companies taking positions in model and AI-infrastructure startups .
The claims that still need outside validation
The round gives Mirendil financial firepower, but the public evidence for its technical claims is still narrow. TechTimes noted that Mirendil’s release names no valuation and relies on company-reported tests rather than independent model evaluations . That distinction is essential: a billion-dollar round can validate investor appetite, but it does not by itself validate performance, reliability or commercial readiness.
Mirendil says agents wrote more than 99% of the code behind its Mirendil-Base training run, while a human researcher set direction and judged results . TechTimes emphasized that this is AI-assisted engineering with a human in charge, not proof of fully autonomous recursive self-improvement . In other words, the company is presenting a powerful workflow, not claiming that its system has removed researchers from the loop.
There is also a scale question. The company’s sub-$10 million compute figure applies to Mirendil-Base, while SiliconANGLE reports that Fern is built on a version of Mirendil-Base with 100 billion to 500 billion parameters and that Mirendil has previewed larger model variants in the 500 billion to 10 trillion parameter range . TechTimes cautioned that comparing the $1 billion raise with the sub-$10 million training bill can be misleading because the cited compute cost covers the smaller model, not the full model family or future training program .
What the round changes competitively
For rivals, Mirendil’s raise raises the cost of staying in the race. A company with fresh billion-dollar financing can hire researchers, reserve compute, negotiate cloud capacity and run more experiments in parallel. That does not guarantee success, but it changes the tempo of competition.
For investors, the round reinforces a 2026 theme: the most valuable AI startups are not only building applications on top of existing models; some are trying to build the machinery that produces better models. Mirendil’s argument is that the bottleneck is not just inference or user experience, but the research process itself . If agents can run experiments, diagnose failures and iterate across days-long workflows, then the “factory” for new models could become more automated .
For customers, the near-term picture is more cautious. EconoTimes reported that Lor is still in private beta and that Mirendil has not posted general release dates for Lor or Fern . That means enterprises and research organizations should treat the announcement as a financing and roadmap milestone, not as evidence that the tools are broadly available today.
The bottom line
Mirendil’s $1 billion Series A is a major vote of confidence in agent-driven model development. It gives the company the capital to expand research capacity, push Mirendil-Base and Fern further, and develop Lor as a platform for longer-running AI research workflows .
But the next proof point will not be the size of the financing. It will be whether Mirendil can turn company-reported results into independently tested models, reliable products and commercial releases that justify the scale of the round. Until then, the raise marks the moment Mirendil entered billionaire difficulty mode — with all the opportunity, scrutiny and competitive pressure that comes with it.
Comments
Be the first to comment.