Suite of 57 Atari 2600 games. Standard benchmark for deep reinforcement learning agents.
Mean HNS across games. Human baseline = 100. Scores >100 exceed average human performance.
Higher is better
Muted rows were not state of the art when published — an earlier or same-year result already scored better.
| Rank | Model | Trust | Score | Year | Links | Fix |
|---|---|---|---|---|---|---|
| 01 | go-explore | vendor | 40000 | 2026 | Source ↗ | Looks wrong? |
| 02 | agent57 | vendor | 4731.3 | 2026 | Source ↗ | Looks wrong? |
| 03 | MEME | verified | 4087 | 2026 | Source ↗ | Looks wrong? |
| 04 | bbos-1 | vendor | 1100 | 2026 | Source ↗ | Looks wrong? |
| 05 | gdi-h3 | vendor | 950 | 2026 | Source ↗ | Looks wrong? |
| 06 | dreamerv3 | vendor | 840 | 2026 | Source ↗ | Looks wrong? |
| 07 | muzero | vendor | 731 | 2026 | Source ↗ | Looks wrong? |
| 08 | EfficientZero V2 | vendor | 242.8 | 2026 | Source ↗ | Looks wrong? |
| 09 | rainbow-dqn | vendor | 231 | 2026 | Source ↗ | Looks wrong? |
| 10 | BBF (Bigger, Better, Faster) | vendor | 224.7 | 2026 | Source ↗ | Looks wrong? |
| 11 | DIAMOND | vendor | 145.9 | 2026 | Source ↗ | Looks wrong? |
| 12 | STORM | vendor | 126.7 | 2026 | Source ↗ | Looks wrong? |
| 13 | Simulus | vendor | 110 | 2026 | Source ↗ | Looks wrong? |
| 14 | DART | vendor | 102.2 | 2026 | Source ↗ | Looks wrong? |
| 15 | human-gamer | vendor | 100 | 2026 | Source ↗ | Looks wrong? |
| 16 | dqn | vendor | 79 | 2026 | Source ↗ | Looks wrong? |