Truth Check/7 October 2026
“"722 manuscripts organized into 372 families", ~4,000 problems posed, three hours of ChatGPT Pro compute per result on average”
The claim as it travelled: OpenAI, openai/math README, 6 Oct 2026.
Holds
The claim holds against the page that produced it.
- Checked Against
- github.com ↗Primary
- Read the Check
- A trillion-parameter model for $1.36 a million tokens, open-weight by month-end, that does the cyber work Claude and GPT refuse. OpenAI's 722 maths papers: 4,000 problems in, not all checked.
Mistral's open weights against Anthropic's three vetted tiers - two answers, one day, to who gets the model that proves a flaw is real; "OpenAI's AI produced 722 new mathematical results" against the repository's own README; and the record check, one prompt that tells you whether your agent can edit its own history.
Every claim we have checked, in the Ledger.
Every claim checked against the page that produced it.
Three mornings a week, free.