TL;DR
- Mirendil raised $1 billion in a Series A — a first priced round the size of an IPO — announced Thursday, November 12, 2026, and co-led by Kleiner Perkins and Clearlake Capital, with Battery Ventures, Doerr Capital, Gil & Co, Google, M13, Pear VC and S32 among the participants (Mirendil company announcement, November 12, 2026; Crunchbase News, November 13, 2026).
- The company builds agents that do research itself. Its Lor platform runs AI agents that write code, run experiments and provision compute for days at a time, with the stated aim of letting one person do work that previously required a team (Mirendil, October 28, 2026).
- The round landed less than five months after a $200 million seed at a $1 billion valuation in June — total disclosed funding is now about $1.2 billion (Unite.AI; TechTimes, November 13, 2026).
- The valuation for this round was not disclosed. Reports of September fundraising talks near $5 billion are unconfirmed talk, not final terms — no source confirms them, and this article does not present them as fact.
Watch the breakdown in about 40 seconds: Episode 28 of The White Report Startup Series — find the reel on Instagram (@acevvswhite) and TikTok (@TheWhiteReport).
The round: a billion dollars, at Series A
Series A rounds are supposed to be the small ones — the first priced check after a seed, sized to prove a product. Mirendil's is $1 billion. The San Francisco startup announced the financing on Thursday, November 12, 2026, in a post published at 9 a.m. Pacific (DailyMech, November 13, 2026). Kleiner Perkins and Clearlake Capital co-led; law firm Sidley Austin confirmed it represented Clearlake on the co-lead investment (Sidley Austin, November 13, 2026). Crunchbase News ranked it the largest U.S. funding round of its week, ahead of every other AI financing in the November 7–13 window (Crunchbase News, November 13, 2026).
The speed is part of the story. Mirendil emerged in June 2026 with a $200 million seed round co-led by Andreessen Horowitz and Kleiner Perkins at a $1 billion valuation, with NVIDIA investing afterwards (Unite.AI). The Series A arrived 141 days later (TechTimes, November 13, 2026). Kleiner Perkins has now led both rounds; Google sits on the Series A investor list while also supplying the compute, under a multi-year Google Cloud partnership announced August 6, 2026 (Unite.AI).
What the announcement does not contain is a valuation. The Wall Street Journal's venture newsletter states it plainly: "The company didn't disclose the valuation at which it raised the funding" (WSJ, November 2026). A $5 billion figure circulates in coverage — it traces to Bloomberg's September report of fundraising talks at that level, and to unnamed sources in the Journal's newsletter. Talks are not terms. Until Mirendil or its leads confirm a number, the honest sentence is the one the company wrote: $1 billion raised, valuation not disclosed.
What Mirendil actually sells: agents that do the research
Most agent startups sell a better interface to a chatbot. Mirendil sells the labour underneath research itself. The company was founded by former Anthropic researchers Behnam Neyshabur and Harsh Mehta, with a founding team of about 20 researchers and engineers drawn from Anthropic, xAI, Google DeepMind and OpenAI (Unite.AI; DailyMech, November 13, 2026). Its stated mission is leverage: one person taking on work that required a team, a team taking on work that required an institution.
The product is Lor, introduced October 28, 2026 and still in private, invitation-only beta. Lor's agents run in the cloud and work on machines the user selects — writing code, running experiments, provisioning their own compute, and iterating for days at a time (Mirendil via Unite.AI and Securities.io, November 2026). Underneath it sit the company's own models: Mirendil-Base, a coding model, and Fern 0.1, post-trained from it, which the company says holds up to 10 million tokens of context — enough, in its telling, to keep days of an agent's work in view at once.
The proof point — and its label
Three weeks before the raise, Mirendil published the result the round is really priced on. In an October 21, 2026 research disclosure, the company said a single researcher, directing AI agents, pretrained Mirendil-Base from scratch in three and a half months using less than $10 million in compute — with agents writing more than 99% of the pretraining code.
Read that claim with its label on: it is company-reported. No independent evaluation of Mirendil-Base or Fern appears in any source we reviewed (DailyMech, November 13, 2026; TechTimes, November 13, 2026), the two larger base models were still in training when the report was published, and the company has disclosed no revenue or post-beta pricing. A billion dollars does not verify the claim. It finances the attempt to prove it at scale — and it tells you what the smartest money in the room currently believes is worth financing.
Why it matters for the AI × money audience
Follow the size, not the slogans. A $1 billion Series A is not a bet on a chatbot feature; it is institutional capital — a venture firm and a private-equity firm co-leading — deciding that the scarce asset of the next AI cycle is not another model, but the automation of research itself. If agents can do the hands-on work of frontier science, the cost curve of discovery bends for every lab, drug pipeline and engineering team downstream. If they cannot, this round becomes the most expensive Series A lesson of the cycle. Either way, the signal for anyone allocating attention or capital is the same: the largest checks have moved upstream, from agents that talk to agents that do science.
Sources, all retrieved November 15, 2026: Mirendil company announcement (November 12, 2026) via Unite.AI, Securities.io and DailyMech; Crunchbase News (November 13, 2026); Wall Street Journal venture newsletter (November 2026); TechTimes (November 13, 2026); Sidley Austin (November 13, 2026); TechStartups (November 13, 2026); AI Affairs (November 13, 2026).
The White Report — Where Super Intelligence Meets Money.