Mirendil Raises $200 Million to Accelerate AI Research for Science

By the Editorial Team · Thursday, June 25, 2026

AI startup Mirendil raised $200 million in seed funding led by Andreessen Horowitz, Kleiner Perkins, and Nvidia, valuing the company at $1 billion as it develops tools for recursive self-improvement in AI. This approach aims to democratize advanced AI research capabilities amid industry restrictions by established labs, potentially accelerating scientific innovation in fields like medicine and materials science.

Mirendil, a new artificial intelligence startup, announced on Wednesday that it has secured $200 million in seed funding, valuing the company at $1 billion. The financing round was led by venture capital firms Andreessen Horowitz and Kleiner Perkins, alongside technology giant Nvidia. The San Francisco-based company aims to develop AI tools that accelerate AI research itself, with the goal of democratizing access to advanced AI capabilities.

Founded by Behnam Neyshabur and Harsh Mehta, former researchers at Google and later Anthropic, Mirendil plans to make AI systems capable of iterative self-improvement—commonly referred to as recursive self-improvement—to help scientists and developers build specialized AI models in fields such as medicine and materials science. “What we are doing is kind of AI for AI for science, as opposed to AI for science,” said Neyshabur, Mirendil’s chief executive, highlighting the company’s focus on creating AI that enhances the development of further AI applications.

Both Neyshabur and Mehta originally connected at Google in 2019, sharing an interest in advancing AI for scientific discovery. They transitioned to Anthropic in late 2024 and departed in December 2025 after the launch of Anthropic’s Claude Opus 4.5 model, which significantly expanded AI agents’ task complexity. Currently, Mirendil employs approximately 20 technical staff and plans to release a model and product in the coming months to gather user feedback.

The company’s launch and significant funding comes amid ongoing tensions within the AI industry over access to powerful models. Leading AI labs like Anthropic use AI to accelerate their own research but restrict external developers from leveraging these tools similarly. Anthropic, for example, prohibits customers from using its AI to develop competitive products, citing national security concerns and the need to preserve the U.S. lead in advanced AI technologies. After imposing export controls in the U.S., Anthropic also suspended access to some of its advanced models, including Fable 5 and Mythos 5.

While some experts and AI-safety researchers warn of the risks posed by self-improving AI—specifically the potential for rapid capability growth beyond human oversight—Neyshabur and Mehta maintain that these concerns are manageable challenges rather than insurmountable barriers. Neyshabur acknowledged the difficulties inherent in safely supervising recursive self-improvement but rejected outright skepticism about its feasibility.

Andreessen Horowitz investor Matt Bornstein described established AI labs as “rational economic actors” in withholding access to their models’ acceleration capabilities, underscoring the market opportunity for independent companies like Mirendil to fill this gap. Along with the two founders, the team includes Shayan Salehian, an early member of xAI, and Tara Rezaei, a Massachusetts Institute of Technology graduate.

The startup’s name, Mirendil, draws inspiration from Tolkien’s Elvish language and translates roughly to “friend of precious things,” reflecting its mission to be a valuable partner in advancing scientific AI research.

This article was reported by synthesizing coverage from multiple independent news outlets to present a balanced, unbiased account. Read about our editorial standards. Spotted an error? Let us know.