Robocurve raises $10M to test robot AI claims independently
The open-source evaluation harness has 97,000+ downloads in under three months, and researchers from 200+ institutions have signed up to build benchmarks.
Jay Chooi says Robocurve has raised a $10M seed round to evaluate frontier AI in the physical world. The company is set up as an independent Public Benefit Corporation, and the round is led by Initialized, with participation from Notable Capital, Decasonic, Y Combinator, Halcyon Futures and others.
The pitch is third-party scoring. Robotics companies grade their own homework almost by default, and a funded outside evaluator is rare enough that the setup is the news here, not any individual result.
The numbers so far
Chooi says Robocurve's research has been viewed more than 6 million times in under three months, and its evaluation harness has been downloaded more than 97,000 times in that same window. Researchers from more than 200 institutions have signed up to build benchmarks with the company, including 19 of the world's top 20 universities by his count.
The harness is fully open source, and Chooi says Robocurve publishes benchmarks that anyone can run and verify. That matters more than the download count. A benchmark you cannot rerun is a press release. A benchmark with the harness attached is something a skeptic can attack on their own hardware.
Shown versus claimed
No independently scored model results are in the announcement. The downloads, the views and the institution count are the only numbers on the table, and they are all self-reported by Chooi. Whether Robocurve's benchmarks become the thing labs actually get judged on, or another leaderboard that vendors quote when it flatters them, depends entirely on what it publishes next.
The timing is easy to read. Robot foundation model launches keep arriving with in-house evaluation numbers and no outside check, and some of the most-shared benchmark threads in this field come from accounts nobody can identify. Chooi himself has been part of that flow, posting an eval that put GPT-6 Astra at 95% on a robot arm control task. Funding a standing evaluator is the structural fix for that, assuming the evaluator stays independent of the labs it scores.
Chooi's closing line is an open invitation, both to companies that want their systems tested and to anyone who wants to compete with him on the testing side. "We welcome more independent evaluators," he wrote. That is an unusual thing for a freshly funded startup to say about its own market, and it is the right thing to say if the goal is a field where claims can be checked.
The more third-party evaluators, the better society can understand how fast robotics AI is progressing.
Our evaluation harness is fully open-source, and we publish benchmarks anyone can run and verify.