Mirendil is in talks for a new funding round at a $5 billion valuation, Bloomberg reports, citing people familiar with the matter. Kleiner Perkins is said to be leading it with up to $1 billion in new capital, and Andreessen Horowitz is in discussions to join.
The company was founded by former Anthropic researchers. Two of the cofounders are Behnam Neyshabur and Harsh Mehta, who left Anthropic in January 2026.
Three months, five times the price
Three months ago, a $200 million seed round became public, also led by Kleiner Perkins and Andreessen Horowitz. It valued Mirendil at $1 billion. That was the company’s first financing of any kind.
Now, one quarter later, the figure on the table is five times that, with the same two investors back at it. The round is not closed; Bloomberg’s sources describe talks.
Self-improving AI, built for other people
Mirendil builds recursively self-improving AI systems, and it aims them squarely at open-source developers and independent labs. The stated goal is to make frontier research accessible enough that people in drug discovery, chemistry or biology can use it without turning themselves into a frontier lab. The company sits in downtown San Francisco with around 20 researchers.
That subject happens to be under political scrutiny right now. Two days ago I wrote that OpenAI wants global standards for self-improving AI, and the UN science panel wants agents slowed down on precautionary grounds. Mirendil is heading the other way and wants to spread the technology around.
Anthropic is the best résumé in the valley right now
The story fits a pattern. Anthropic is not just the employer whose models everyone benchmarks against; it is also the address from which money is easiest to raise. Two names on a résumé have been worth a $4 billion jump in valuation in three months, with no public product.
So the direction matters more here than the sum. While Washington and the UN argue about brakes for self-improving systems, two of the best-known venture firms are funding a company whose plan is to put exactly those systems into as many hands as possible.