Rumor Claims Claude Opus 5 Achieves Perfect 2026 IMO Score Without Tools

Social media is buzzing with unverified claims that Anthropic's next-generation model, Claude Opus 5, achieved a perfect 42/42 score on the 2026 International Mathematical Olympiad (IMO). This rumor has sparked widespread discussion within the AI community regarding the mathematical reasoning capabilities of large language models and the validity of current benchmarks.

Confirmed

What can be confirmed is the circulation of information on social platforms: a viral evaluation report screenshot claims Claude Opus 5 scored 42/42. According to the posts, the test was conducted without using an agent harness or any external tools. The model generated four independent solutions for each of the 6 problems, which were then graded by three evaluators. Furthermore, poster @davidwromero noted that 6 of the solutions were judged correct by GPT 5.6, providing the original reasoning traces.

Unconfirmed

The authenticity and authority of the evaluation results cannot be confirmed at this time. The 2026 IMO has not yet taken place, indicating that the test likely used leaked, simulated, or internally generated questions. Additionally, the organizers of the evaluation, the identities of the three evaluators, and the rigor of the grading process remain unclear, and the use of models like GPT 5.6 as a grading standard lacks official endorsement.

Why it matters

If the test results are accurate, it would represent a dominant performance by an AI model on highly difficult, strongly filtered math problems. Commenters like @surmenok argue that this indicates existing high-difficulty math benchmarks are now "completely saturated," and that more challenging evaluation systems will need to be designed in the future to measure the true upper limits of AI capabilities.

2026-07-25 ~ 2026-07-26 · 5 related posts

Primary sources

1 near-duplicate retellings: surmenok