Learn2Play Bench - Mean: leaderboard

Learn2Play Bench's average score over all scored episodes (5 per normal game, 10 per challenge game) rather than each trial's peak, so a model that needs many failed episodes to learn a game's hidden rules scores lower here than on Max; same 20 games, OpenCode harness and reference-ceiling normalization.

Metric: Normalized Mean (% of each game's reference ceiling, challenge games weighted 3). Source: liushiliushi.github.io. Saturation forecast: Around February 2028. 11 models tracked. Updated 19 Nov 2026.

Top models

#ModelSettingScoreOverall rank
1Claude Opus 5.561#9
2GPT-6 AstraMedium56.1#10
3Claude Opus 553.8#26
4Gemini 3.5 Flash52.2#108
5Qwen 3.8 Max51.1#66
6Kimi K350.2#61
7GPT-5.6 Sol41.1#36
8Gemini 3.7 Flash39.9#42
9DeepSeek V4 Pro38.2#198
10Claude Sonnet 535.6#118
11GPT-5 Mini22.3#365

No result here: #1 Claude Fable 5.2, #3 Gemini 4 Neon, #4 GPT-6.2 Astra.

Interactive version: theaggregate.ai/benchmark?slug=learn2play-bench-mean · How It Works · Data refreshed daily, snapshot 2026-11-20.