OpenAI internal model solves over 100 math problems in 24 days

The decision to convene an independent advisory group is the most telling detail here: OpenAI is treating verification, not raw capability, as the bottleneck, which suggests the claims are extraordinary enough that internal validation is not being trusted. That matters for mathematicians whose fields are named as affected, and for anyone assessing how quickly AI is moving from assisting proofs to generating results that require peer review. Whether the advisory group's findings are published, and whether the solved problems are independently reproduced, is the open question.