Text / LLM / Benchmark
Recent LLM benchmarks
Commonly reported LLM benchmarks; grouping used on this site, not an official suite. Harder reasoning, math, science and coding benchmarks typical of 2024-2026 model reports.
Test split: GPQA Diamond: all 198 questions (no train/test split). MMLU-Pro: test (12,032; the 70-question validation split supplies few-shot examples). LiveCodeBench: code-generation problems inside a release-date window (this site: release_v6, May 2023 - Apr 2025, 1,055 problems; reports often use a later sub-window). SWE-bench Verified: the 500-instance test set. AIME 2024 / AIME 2025: all 30 problems of AIME I and II. MATH-500: the 500-problem test.jsonl from PRM800K.
Contamination matrix
- TestTest data seen17
- SeenSeen, split unclear0
- TrainTraining data seen0
- Related data seen10
- NoNot seen65
- ?Unknown538
Hover a cell for the reason; click it for the evidence.
Notes
Membership chosen by this site.
Sources
- GPQA paper · paper · retrieved 2026-10-15
- MMLU-Pro paper · paper · retrieved 2026-10-15
- TIGER-Lab/MMLU-Pro dataset card · dataset card · retrieved 2026-10-15
- LiveCodeBench README · readme · retrieved 2026-10-15
- Introducing SWE-bench Verified · website · retrieved 2026-10-15
- SWE-bench Verified dataset card · dataset card · retrieved 2026-10-15
- 2024 AIME I - AoPS Wiki · website · retrieved 2026-10-15
- 2025 AIME I - AoPS Wiki · website · retrieved 2026-10-15
- openai/prm800k README (MATH splits) · readme · retrieved 2026-10-15