Mirendil raises $200M seed at $1B valuation to automate AI research

Mirendil raised a $200 million seed round at a $1 billion valuation on June 24, 2026, to build frontier AI systems that automate scientific and engineering research and development. The round, co-led by Andreessen Horowitz and Kleiner Perkins with Nvidia as a co-investor, ranks among the largest seed rounds in AI history.

The round stands out as much for its team as its size. The startup was founded by former Anthropic and Google researchers, including Behnam Neyshabur, who spent years at Alphabet co-leading Gemini’s reasoning research before joining Anthropic, and Harsh Mehta, who built Anthropic’s internal autoresearch platform. Two other founders came from xAI and MIT.

Mirendil is building models meant to run the research loop itself: designing experiments, interpreting results and managing compute allocation, according to Andreessen Horowitz’s announcement. The company is targeting materials science, drug discovery, chemistry, biology and robotics as application domains.

Andreessen Horowitz argued in its post that the full potential of AI will not be realized until the technology is placed in the hands of builders, framing Mirendil as the company putting frontier AI research within reach of engineers and scientists outside the big labs.

The disclosure carries clear limits. Mirendil reported no revenue and named no product ship date at launch, and the description of “one of the largest” seed rounds is the company’s framing rather than a verified record. A startup automating its own research domain also faces an unproven core premise, that models can reliably manage open-ended scientific work, not just discrete tasks.

The next signal will be whether Mirendil ships a usable product or research result against that $1 billion valuation, rather than a pitch.