Benchmark contamination
Part of Evaluation and benchmarking
Benchmark contamination is losing ground in AI research: 0.1% to 0.1% of new AI papers, -0.0 points. It ranks #141 of 214 by gain in share. As of Oct 10, 2026.
- Change in share
- -0.0 pts
- Share, last 3 months
- 0.1%
- Papers, last 3 months
- 33
- Ready for products
- 2.1 / 5
More in Evaluation and benchmarking
Agent benchmark1.4% → 1.6%+0.2 ptsCounterfactual evaluation0.1% → 0.2%+0.1 ptsVision language model evaluation0.2% → 0.3%+0.1 ptsSoftware engineering benchmark0.2% → 0.3%+0.1 ptsLLM as a judge0.3% → 0.4%+0.1 ptsMultilingual evaluation0.2% → 0.3%+0.0 ptsMulti turn dialogue evaluation0.1% → 0.1%+0.0 ptsFaithfulness evaluation0.1% → 0.1%+0.0 ptsLLM safety evaluation0.1% → 0.1%+0.0 ptsVideo generation evaluation0.1% → 0.1%+0.0 pts
Companies working on it
Benchmark contamination: quick answers
No. Its share of new AI papers went from 0.1% to 0.1%, with 33 papers in the last 3 months.