On the 2025 USAMO, none of the leading reasoning LLMs tested scored higher than 5% overall.

Was it true for the period described?

Outcome: Supported

The dated evidence matches what the claim committed to, under the criteria frozen before grading.

Scored on a plain reading. The speaker named no measure for this claim, so it was judged on the reading an informed reader would take, stated below under our interpretation. Both reviews had to find that reading fair.

Evidence as of

2026-10-28

Deadline state when assessed

No horizon

Assessment recorded

2026-10-29

Policy version

2026-10-28.7

Rationale

The original evaluation's highest average total was 2.0 points out of 42, approximately 4.76%. All six models Marcus named were therefore below his 5% threshold.

Limitations

• This describes the original evaluation with its particular prompts, sampling procedure and expert grading.
• It does not establish a ceiling on these models under other methods or on subsequently released models.

Evidence (1 item)

• Supports the claim: Petrov et al., ETH Zurich and INSAIT; arXiv, 2025-03

  four runs

  Why it matters here: Table 1 reports average totals of 2.0, 1.8, 1.5, 1.2, 1.2 and 0.9 points for the six named models, against a maximum of 42. The highest reported total therefore corresponds to approximately 4.76%, below 5%.

  Where

  Table 1, caption and Total column

  Publisher

  Petrov et al., ETH Zurich and INSAIT; arXiv

  Kind of source

  Preprint, not peer reviewed

  Arithmetic

  100 × max(2.0, 1.8, 1.5, 1.2, 1.2, 0.9) / 42 (inputs: Table 1 displayed totals: R1 2.0; Flash-Thinking 1.8; Claude 3.7 1.5; QwQ 1.2; o1-pro 1.2; o3-mini 0.9; Maximum total: 42 points; Approximately 4.76%, calculated from the displayed totals. Their rounding does not put the maximum above 5%.)

  Describes events of

  2025-03

  Public from

  2025-03-27