Mirendil Raises $1 Billion Series A on a One-Researcher Model Bet

Mirendil has raised a $1 billion Series A led by Kleiner Perkins and Clearlake Capital, betting that AI agents (autonomous software systems that plan and execute tasks) can do the hands-on work of an entire research lab. The round, announced Thursday, values the San Francisco startup at roughly $5 billion according to an Andreessen Horowitz talent newsletter — five times its seed valuation from barely five months ago. Mirendil did not disclose a valuation itself.

The wager rests on a striking claim. Mirendil says one researcher, working with a fleet of agents and less than $10 million in compute and token spend, built its first foundation model from an empty repository in 3.5 months. For context, the company argues that pretraining a competitive model has historically taken dozens of specialists at open-model efforts and hundreds at frontier labs. "When agents do the hands-on work, the size of your team stops limiting what you can build," Mirendil wrote in its announcement.

Those claims come from the company itself, and no third party has verified the compute figure or the single-researcher account. That caveat matters: a $1 billion check is buying the narrative before the results have faced independent scrutiny.

Who put in the money?

Battery Ventures, Doerr Capital, Gil & Co, Google, M13, Pear VC and S32 joined Kleiner Perkins and Clearlake in the round. Google's participation extends an existing relationship: in August, Mirendil signed a Google Cloud partnership worth more than $100 million for a mix of TPUs (Google's custom AI chips) and Nvidia GPUs, according to the a16z newsletter.

The funding timeline is unusually compressed. Mirendil came out of stealth on June 24 with a $200 million seed round led by Andreessen Horowitz and Kleiner Perkins, followed by an Nvidia investment; SiliconANGLE reported that seed valued the company at $1 billion. On Sept. 22, Bloomberg reported, citing people familiar with the matter, that Mirendil was in talks to raise up to $1 billion at a $5 billion valuation including the new money. The Series A closed those talks in under seven weeks.

The takeaway: a five-fold valuation step-up in five months, in a market where most seed-stage companies wait years for a Series A.

Who is behind Mirendil?

The founders come from the labs Mirendil now wants to undercut. Chief Executive Behnam Neyshabur co-led Google DeepMind's Blueshift team, which worked on reasoning and math for Gemini. Chief Technology Officer Harsh Mehta built the first version of Anthropic's internal AI research platform.

Both joined Anthropic in late 2024 and left together in December 2025 to found the company, according to Dealroom. The a16z post describes a team of about 20 researchers and engineers — a headcount that would make the $5 billion valuation roughly $250 million per employee. The company has not disclosed burn rate or revenue.

What did one researcher actually build?

The headline artifact is Mirendil-Base, a coding-focused model in the 100 billion to 500 billion parameter range. In an Oct. 21 technical post, Mirendil said a single researcher built it in 3.5 months, with agents writing more than 99% of the code, running the experiments and diagnosing failures while the researcher decided what to try and what to trust.

The company says the sub-$10 million figure is all-in, covering the agents' token spend and data processing. It says the model used only publicly available data, with no distillation (extracting outputs from a larger model to train a smaller one) and no licensed datasets. Two larger versions are in training: one with 500 billion to 2 trillion parameters and another with 2 trillion to 10 trillion. Mirendil has not published benchmark comparisons against frontier competitors, nor said when the larger models will finish training.

The takeaway: the parameter range is frontier-adjacent, but without independent benchmarks the "competitive" claim rests on Mirendil's own framing.

What are Fern and Lor?

Mirendil has layered two products on top of the base model. Fern is a post-trained variant that holds up to 10 million tokens of context — roughly half a million lines of code, or days of an agent's work, in view at once. The company pitches it at the routine work of model building: debugging a training run from its logs, or scanning a pretraining shard for duplicates and benchmark contamination. Fern is taking early-access requests now.

Lor is the commercial centerpiece: an invite-only private beta agent platform whose agents write code, run experiments and iterate for days, provision their own GPU nodes and release them when done, and split a job across hundreds of parallel workers. Mirendil says it is building Lor "to do the work of an entire AI lab." Lor runs Anthropic's and OpenAI's frontier models alongside open-source models and Mirendil's own — meaning the startup is simultaneously a customer and a competitor of its founders' former employer.

The company offered two illustrative examples:

• 102 agents built training environments from 100 PyTorch pull requests, averaging eight minutes each.
• Agents found security bugs in LiteLLM overnight and fixed them by morning.

These are company-provided anecdotes, not audited results. Lor is free during the beta, and Mirendil has not said what it will cost afterward or when it will open beyond the waitlist.

Why does a $1 billion round matter now?

The round places Mirendil among the best-funded of the new research labs, a group Dealroom said includes Safe Superintelligence, Thinking Machines Lab and Recursive Superintelligence. It also lands roughly a week before Anthropic — where much of Mirendil's founding team worked — is set to list on Nasdaq after setting its IPO price range.

The stakes are structural. If one researcher plus agents can genuinely pretrain a frontier-scale model for under $10 million, the moat around large labs narrows from compute and headcount to taste and trust. If the claim does not hold up at the 2-trillion-to-10-trillion-parameter scale, Mirendil becomes another well-capitalized bet on a productivity story that investors have heard before.

The company's own roadmap supplies the test: two larger models are in training, and their results — whenever Mirendil chooses to publish them — will show whether agent-led research scales or stalls.