OpenAI is facing questions on two fronts this week, as leading mathematicians challenge how it released hundreds of claimed solutions to difficult math problems, and new reporting suggests its annualized revenue is around $20 billion lower than previously reported.
The lab released 719 manuscripts presenting solutions to some of the hardest open problems in mathematics, saying it had consulted an advisory group of elite mathematicians to avoid the controversy that followed a previous announcement. That body, the Advisory Group on Mathematics and Artificial Intelligence (AGMAI), is hosted by the Institute for Advanced Study and comprises nine prominent researchers. It issued guidelines for frontier labs at the end of September, and said in a statement that it was ultimately up to the mathematical community to judge how well its recommendations had been followed.
OpenAI met some of those principles, including releasing results quickly and explaining how its models reached conclusions. However, the group’s first request was for labs to stop testing advanced problems on proprietary models, which OpenAI’s release explicitly does. Only 10 of the 719 manuscripts included the model’s chain of thought, 42% of the proofs had not been formalized, and the lab did not provide the machine-readable metadata linking natural language and formal proofs that AGMAI had requested. The group also suggested OpenAI help fund human mathematicians whose work is needed to make the results meaningful.
Mathematician Terence Tao wrote on social media that problems are being solved autonomously by AI prompters who lose interest in the field once a target is solved and cannot explain the output well enough to engage with other researchers. Those concerns were reinforced by a paper from mathematicians at the University of Cambridge and King’s College London, which documented at least two discrepancies between the natural language proof and the Lean code behind OpenAI’s solution to a problem derived from the Navier-Stokes equations. The authors concluded that such autoformalized proofs should not be trusted without the same peer review applied to other work. Harvard mathematics professor Melanie Wood told TechCrunch that human understanding of these results does not exist at the point of release, and that the real work begins afterwards.
Meanwhile, the Financial Times reported that OpenAI has told investors its annualized revenue is nearing $50 billion, well below the roughly $70 billion figure reported just over a week ago. According to the FT, that higher number emerged from attempts by OpenAI’s own investors to create a direct comparison with Anthropic, which has reported a $65 billion run rate. The two companies calculate annualized revenue differently, with Anthropic counting sales made through its cloud partners and OpenAI excluding them.
Revenue has become a pressing issue as OpenAI works to justify the enormous investment behind it, including $122 billion secured in a March funding round. Leaked 2025 financials showed the company earned about $13 billion while spending significantly more, and its long-rumored IPO has been pushed back to early 2027.
