Mirendil raises $1B Series A led by Kleiner Perkins and Clearlake
Mirendil, a San Francisco startup, said on November 12, 2026 that it had secured $1 billion in Series A funding, according to its own release distributed through PR Newswire. Kleiner Perkins and Clearlake Capital co-led the round, and Battery Ventures, Doerr Capital, Gil & Co, Google, M13, Pear VC and S32 also took part alongside other new and existing investors. The company has not disclosed a valuation. SiliconANGLE reports that the company was launched last year by two former Google DeepMind and Anthropic researchers, and that chief executive Behnam Neyshabur co-invented SAM, a technique for improving how models handle tasks outside their training.
The round follows a $200 million seed in June 2026 that Andreessen Horowitz and Kleiner Perkins led, with Nvidia joining afterward, according to WebProNews. Mirendil's first product, Lor, is in private beta and splits model training work across hundreds of agents. The company says Lor can run agents that write code and run experiments for days, and that it works with models from Anthropic and OpenAI as well as open-source models and Mirendil's own.
The headline example is a single researcher who pretrained Mirendil-Base, a coding model, from scratch in 3.5 months with less than $10 million in compute. Mirendil then post-trained that model into Fern, which handles up to 10 million tokens of context. These results come from the company and have not been independently replicated. MLQ notes that an experienced researcher directed the agents, so the process was not fully autonomous. Mirendil says the new capital will go toward Mirendil-Base, Fern and Lor.
Why it matters
For founders building AI companies, the round suggests investors will fund an agent-driven research stack at billion-dollar scale before general release. If agents can absorb the engineering and experiment load that once required large research teams, a small company may be able to test model ideas that previously needed a major lab's headcount. The bottleneck could shift from people to compute budgets, evaluation quality and access to frontier models.
For sellers and funders, the claims need checking before they shape plans. The valuation is undisclosed, the 3.5-month pretraining result is the company's own account, and Lor is still gated behind a waitlist with no announced pricing. Operators evaluating similar tools should ask for independent benchmarks, a full accounting of compute and agent costs, and a clear description of how much human supervision each run required.
Key points
- 1Mirendil announced the round on November 12, 2026, with Kleiner Perkins and Clearlake Capital as co-leads.
- 2The company has not disclosed a valuation for the Series A round.
- 3The June 2026 seed raised $200 million, led by Andreessen Horowitz and Kleiner Perkins, with Nvidia joining.
- 4Lor, its first product, is in private beta and free during the beta on macOS, Linux and web.
- 5
