BeyondAIME: Advanced Math Problems for AI Benchmarking

ByteDance-Seed's BeyondAIME is a curated test set for benchmarking advanced mathematical reasoning. The dataset contains problems sourced from high-school and university mathematics competitions, with a difficulty level greater than or equal to that of AIME Problems #11-15. It was last updated on June 17,  2025.

Use Cases

• Benchmarking the performance of large language models on high-difficulty math problems.
• Training models to solve competition-level mathematical reasoning tasks.
• Analyzing the difficulty progression and problem-solving strategies required for advanced math.
• Evaluating the generalization of AI models beyond standard high-school mathematics.

Strengths

• Problems are curated to a high difficulty level, specifically greater than or equal to AIME Problems #11-15.
• Problems are sourced from established high-school and university mathematics competitions.

Limitations

• Column-level documentation is absent; field semantics must be inferred after download.
• Row count is unknown, which may limit suitability assessment.
• Description metadata is limited; actual data quality requires manual inspection after download.

Provenance

Source

ByteDance-Seed

Collection Method

Curated from high-school and university mathematics competitions.

Time Range

null

Freshness

Last updated 2025-06-17 04:35:38; freshness should be verified

Geography

null