CodeContests (10@100k): leaderboard
Competitive programming on the CodeContests test set (Li et al., DeepMind, 2022): 165 Codeforces-style problems published after the training data, judged on hidden and generated tests. Percentage of problems solved when 100,000 programs are sampled per problem, filtered on the example tests and at most ten submitted, the AlphaCode protocol; higher is better.
Metric: Test-set solve rate (%; 10 submissions per problem chosen from 100,000 samples after filtering on the example tests). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 2 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | AlphaCode 41B | 27.7 | |
| 2 | AlphaCode 9B | 25.8 |
Interactive version: theaggregate.ai/benchmark?slug=codecontests-10-100k · How It Works · Data refreshed daily, snapshot 2026-11-12.