ALE-Bench: leaderboard
Score-based algorithmic programming benchmark built from AtCoder Heuristic Contest tasks, evaluating AI systems on hard optimization problems with hidden/private test evaluation.
Metric: Performance (Self-Refine x1) (self-reported). Source: benchmarklist.com. Status: saturation imminent. 89 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 (xHigh) | 1942.97 |
| 2 | GPT-5.3 Codex (xHigh) | 1655.22 |
| 3 | GPT-5.4 (High) | 1607 |
| 4 | GPT-5.4 (Medium) | 1520.72 |
| 5 | Gemini 3 Flash (Preview) (High) | 1367.2 |
| 6 | Claude Sonnet 4.6 (Medium) | 1327.3 |
| 7 | Claude Opus 4.7 (Thinking) | 1323.05 |
| 8 | GPT-5.2 Codex (xHigh) | 1299.9 |
| 9 | GPT-5.2 (High) | 1293.55 |
| 10 | GPT-5.2 (Medium) | 1249.83 |
| 11 | GPT-5.1 Codex (High) | 1244.92 |
| 12 | GPT-5.1 (Thinking) | 1192.15 |
| 13 | GPT-5.4 Mini (High) | 1188.58 |
| 14 | Gemini 3 Pro (Preview) (High) | 1176.75 |
| 15 | GPT-5 (Thinking) | 1162.45 |
Interactive version: theaggregate.ai/benchmark?slug=ale-bench · How It Works · Data refreshed daily, snapshot 2026-09-05.