Former Anthropic Researchers Launch Mirendil at $1 Billion Valuation With $200M Seed Round

Former Anthropic researchers have launched Mirendil with an ambitious objective: developing artificial intelligence systems capable of accelerating AI research itself. The San Francisco-based startup has emerged from stealth with a $200 million seed funding round at a $1 billion valuation, one of the largest seed financings ever announced by an AI startup. The round was co-led by Andreessen Horowitz and Kleiner Perkins, with participation from NVIDIA.

Founded by Behnam Neyshabur and Harsh Mehta, who previously worked together at both Google and Anthropic, Mirendil believes that the next leap in artificial intelligence will come from systems designed to help scientists and engineers build better AI models rather than simply serving as end-user assistants.

According to the company, its long-term vision is to democratize frontier AI research by making advanced research capabilities available beyond a handful of well-funded AI laboratories.

Building AI for AI Research

Rather than focusing on consumer chatbots or enterprise productivity software, Mirendil describes itself as a frontier AI laboratory developing systems that excel at AI research and development.

The company’s thesis is that many scientific breakthroughs begin with researchers building deep expertise over long periods of time. Mirendil aims to create AI systems capable of developing and accumulating that expertise, allowing scientists to accelerate research across specialized domains.

Instead of simply generating answers, these systems are intended to assist researchers throughout the scientific process, helping develop increasingly capable AI models that can then be adapted for disciplines such as biology, chemistry, medicine, materials science, and robotics.

Democratizing Frontier AI

One of Mirendil’s core arguments is that today’s frontier AI development has become concentrated within a relatively small number of organizations possessing enormous computational resources.

The company believes that many universities, startups, pharmaceutical companies, research institutes, and industrial laboratories could make faster scientific progress if they had access to AI systems specifically designed to accelerate research rather than general-purpose language models.

Instead of replacing scientists, Mirendil envisions AI acting as a research partner that helps experts iterate faster, explore more hypotheses, and develop increasingly specialized models for their own fields.

A Team Drawn From Leading AI Labs

Mirendil’s founding team reflects the growing migration of experienced researchers from major AI laboratories into independent startups.

CEO Behnam Neyshabur previously conducted research at Google before joining Anthropic, while CTO Harsh Mehta also worked as a research scientist at Anthropic. The broader founding team includes Shayan Salehian, an early member of xAI, and Tara Rezaei, an MIT graduate. Together, the company has assembled roughly 20 researchers and engineers with backgrounds spanning Anthropic, OpenAI, Google DeepMind, and xAI.

A Different Approach to AI Infrastructure

Mirendil’s website emphasizes that its mission extends beyond building another foundation model. Instead, the company aims to redesign the research process itself by developing systems that help engineers and scientists perform AI research more effectively.

The startup says it is building frontier models alongside new research workflows intended to create self-improving AI systems that continuously enhance scientific and technological development. The company believes this approach can make advanced AI research accessible to organizations working on problems ranging from drug discovery and chemistry to biology and robotics.

The company also plans to release its first model and product in the coming months to gather feedback from researchers and developers as it refines its technology.

Recursive Self-Improvement Remains a Controversial Area

Mirendil’s vision intersects with one of the most debated concepts in artificial intelligence: recursive self-improvement, where AI systems contribute to building increasingly capable successors.

Supporters argue that such systems could dramatically accelerate scientific discovery by shortening research cycles and allowing experts to focus on higher-level reasoning. Critics, including some AI safety researchers, have warned that increasingly autonomous AI development raises important questions around oversight, control, and governance.

Mirendil’s founders argue that these challenges are difficult engineering problems rather than insurmountable barriers, maintaining that sufficiently robust supervision can allow recursive improvement to be developed safely.

What the Funding Means for the Future of AI Research

Mirendil’s unusually large seed round highlights growing interest in AI infrastructure designed specifically for research rather than consumer applications. As frontier AI systems become increasingly capable, demand is likely to grow for tools that can help scientists and engineers build, evaluate, and refine specialized models more efficiently.

If AI systems become better at supporting AI research itself, organizations outside the largest frontier labs—including universities, pharmaceutical companies, healthcare providers, industrial manufacturers, and research institutes—could gain access to capabilities that were previously limited to a handful of well-funded organizations. Rather than concentrating AI development within a small number of companies, technologies like those being developed by Mirendil could enable a broader ecosystem of domain-specific AI models tailored to specialized scientific and industrial challenges. While significant technical and safety hurdles remain, the emergence of companies focused on AI-assisted AI development signals that the next phase of innovation may increasingly center on accelerating the pace of scientific discovery itself.