Funding & VC

Mirendil Raises $1 Billion Series A on Unverified AI Agent Claims

Mirendil's $1 billion round, led by Kleiner Perkins and Clearlake, prices an 11-month-old lab at a reported $5 billion on internal benchmarks alone. No independent evaluation exists yet.

By Olivia Hart

4 min read

Anthropic Alumni Startup Mirendil Raises $1 Billion Series A on Agent-Driven Model Claims
Anthropic Alumni Startup Mirendil Raises $1 Billion Series A on Agent-Driven Model ClaimsStartupStockPhotos / pixabay

What's News

  • Mirendil raised a $1 billion Series A on November 12, led by Kleiner Perkins and Clearlake Capital.
  • Press reports put the valuation near $5 billion, roughly 5x the reported seed valuation; Mirendil has not confirmed it.
  • The company claims one researcher pretrained a 100–500B-parameter model in 3.5 months for under $10 million, with agents writing over 99% of the code.
  • No independent evaluations of any Mirendil product have been published.
  • Crunchbase News counted 27 venture rounds of $1 billion or more in Q3, about a third of the quarter's $159 billion in global venture funding.

Mirendil, an 11-month-old San Francisco AI lab, has raised a $1 billion Series A led by Kleiner Perkins and Clearlake Capital, with participation from Google, Battery Ventures, Doerr Capital, Gil & Co, M13, Pear VC and S32. The company disclosed the round on November 12. It has not confirmed its valuation, but press reports from September put the target near $5 billion — roughly five times the reported seed valuation from June 2026, when Mirendil raised about $200 million led by Andreessen Horowitz.

Who is behind Mirendil?

Founders Behnam Neyshabur (CEO) and Harsh Mehta (CTO) met at Google in 2019 and later worked together at Anthropic, according to the company's materials. Neyshabur co-led reasoning research for Gemini at Google and ran a research team at Anthropic; Mehta worked on Gemini pretraining and led automated AI research at Anthropic. Early team members include Shayan Salehian, formerly of xAI, and Tara Rezaei, an MIT graduate and former OpenAI student researcher. The bench is thin on paper: earlier reports put headcount at roughly 20 researchers and engineers, and the company calls itself "still a small team with a singular focus."

What has the company actually shipped?

Three artifacts, with every performance claim coming from Mirendil itself:

  • Mirendil-Base: a pretrained coding model of 100–500 billion parameters, published October 21. The company says one researcher, working with agents, pretrained it from scratch in 3.5 months for under $10 million in compute — with agents writing more than 99% of the code.
  • Fern: a post-trained version of Mirendil-Base with a claimed 10-million-token context window, roughly half a million lines of code. It is in early access, and only a basic long-context test has been published.
  • Lor: the agent platform at the core of the pitch. It runs hundreds of research agents that write code and run experiments for days at a time, entered private beta October 28, and, per Mirendil, the company "runs on Lor every day."

Two larger models are still in the oven: Mirendil-Base-Large (0.5–2 trillion parameters) was reported 80% through pretraining as of October 21, and Mirendil-Base-XL (2–10 trillion parameters) 19% through. Andreessen Horowitz has labeled the whole approach "vibe research."

How solid are the benchmark claims?

Mirendil compared Mirendil-Base against several open models on held-out code using "bits per byte," a measure of next-token prediction accuracy where lower is better. The comparison is not like-for-like, because some reference models are post-trained. Bits per byte is also a narrow metric that says little about reasoning or real coding tasks. The company itself concedes it does not think it has reached fully autonomous recursive self-improvement yet. No independent evaluations of Mirendil-Base, Fern or Lor have been published — which means a reported $5 billion valuation currently rests on internal benchmarks and founder reputations.

Why did investors write the check?

Kleiner Perkins, which co-led the seed, doubled down as Series A co-lead. Its June investment note stated the thesis plainly: "Better models do better research. Better research produces better models. The loop is the product." Clearlake Capital is an unusual Series A lead — the private-equity firm closed a $14.8 billion flagship fund in June focused on "AI-driven change." Google plays both sides of the table: it invests and supplies compute under a multiyear Cloud deal worth over $100 million signed in August, a concentration point worth watching. Notably absent from the published Series A list is Andreessen Horowitz, the seed lead; the press release mentions "other new and existing investors" without naming them.

Is this a market top?

The round fits a 2026 pattern. Crunchbase News counted 27 venture rounds of $1 billion or more in Q3 alone, accounting for roughly a third of the quarter's $159 billion in global venture funding. Mirendil has disclosed no revenue. In plain words: a 20-person lab with no public revenue data just absorbed one of the year's characteristic mega-rounds.

What comes next?

Mirendil says the Series A "lets us go further" toward making frontier AI R&D accessible. The near-term milestones are completing Mirendil-Base-Large and -XL, opening Lor beyond private beta, and — critically — third-party testing of the models and the agent platform. Larger labs are also automating their own research with far more data, talent and compute. Until outside evaluations and any revenue signal arrive, the $1 billion check buys a promise, not a proof.

Source: epyhia.com

Share this article:

More from Olivia Hart

Olivia Hart

Show full bio

Staff writer covering people, launches and day-to-day practice in entrepreneurship, startups and venture funding at Business Bearings.

2,203 articles

Related articles

« Previous articleNext article »