Atari 2600.

Mean HNS 10077.52% at 1B frames on Atari 57. Median HNS 1934% (Agent57-style max-over-training). Breaks 24 human world records. ICLR 2023 Oral.

Median HNS across 57 games. First to beat human baseline on ALL games.

Mean HNS at 1B frames on Atari 57 (human=100). 95% CI: 3723–4445. Reaches human-level on all 57 games within 390M frames.

IQM (interquartile mean) = 13.86 at 200M frames on Atari 57. Metric differs from mean/median HNS in other entries — stored as IQM×100 for scale. Automated RL rule discovery. Nature, Oct 2025.

Model-based optimization.

High sample efficiency.

Mastered Atari with fixed hyperparameters using world models.

Model-based agent planning with learned model.

EfficientZero V2. Mean HNS 242.8%, Median 128.6% on Atari 100k (26 games, 100k steps). Surpasses BBF. Model-based RL with Gumbel search. arXiv Mar 2024.

Median HNS. Combines 7 improvements to DQN.

Median HNS. Combines 7 improvements to DQN.

Bigger, Better, Faster (BBF). Mean HNS 224.7% on Atari 100k (26 games, 100k steps). IQM: 104.5%, Median: 91.7%. Value-based RL with scaled networks. ICML 2023.

DIAMOND (Diffusion World Model). Mean HNS 145.9%, IQM 64.1% on Atari 100k (26 games, 100k steps). Best agent trained entirely within a world model. NeurIPS 2024 Spotlight.

STORM (Stochastic Transformer World Models). Mean HNS 126.7% on Atari 100k (26 games, 100k steps). Transformer-based stochastic world model. arXiv Oct 2023.

Simulus. First planning-free world model to reach human-level IQM and median HNS on Atari 100k (26 games, 100k steps). Superhuman on 13/26 games. Combines intrinsic motivation, prioritized replay, regression-as-classification. arXiv Feb 2025.

DART (Discrete Abstract Representations for Transformer-based learning). Mean HNS 102.2%, Median 79.0%, IQM 57.5% on Atari 100k (26 games, 100k steps). ICML 2024.

Professional human tester baseline.

Professional human tester baseline.

Historical baseline (2015). Median HNS.

Historical baseline (2015). Median HNS.