dig.bench<br>Leaderboard<br>CombinedTier breakdown<br>Basic HarnessOpus 5GPT-5.5Kimi K3Gemini 3.1 ProGLM-5.2DeepSeek V4 FlashDeepSeek V4 ProQwen 3.6 27B

Agentic HarnessOpus 5 + Prime AgentFable 5 + Claude CodeGPT-5.6 Sol + CodexKimi K3 + Kimi CodeGemini 3.1 Pro + PRO-LONG

Win rate10.50

Each game is beaten by at least one human on their first attempt

Difficulty tiers

All games are human beatable, on first attempt, as validated on external human testers. Win rate: each game's wins are averaged over its runs, then those are averaged across the tier's ten games.

About<br>dig.bench is a benchmark of scientific discovery.<br>Each of its 70 games measures whether an agent can experiment to discover that game's own unknown rules. Every game is text-based, which puts it in the natural domain of language models: no visual confounds stand between a model and the discovery, so what dig.bench tests is discovery alone. Humans and frontier models play the same games with access to the same information, and progress is scored by whether the game can be beaten within a limited number of steps.<br>The games come in 7 tiers, depending on their difficulty. No game is easy and they all require effortful play, but humans can make the discoveries necessary to solve even our hardest games, while the best models struggle to beat games in the top tier.<br>Scale70 new interactive games (21 publicly released).<br>What qualities of models do we testTo beat each game an agent must discover the unknown rules and apply them to solve challenges.<br>EvaluationHumans and frontier models play through the same interface: identical game states, identical action sets, identical step budgets.

The platform and equivalent human and model interfaces. marks the available actions; creative mode, an option in some games that allows a player to enter another level where they can experiment without it counting towards the step count; the current game state; game statistics; and action and state history.

Play<br>21 of the 70 games are public. Tiers get increasingly harder for models (1 = easiest, 7 = hardest).<br>tier 1P-1tier 1P-2tier 1P-3tier 2P-4tier 2P-5tier 2P-6tier 3P-7tier 3P-8tier 3P-9tier 4P-10tier 4P-11tier 4P-12tier 5P-13tier 5P-14tier 5P-15tier 6P-16tier 6P-17tier 6P-18tier 7P-19tier 7P-20tier 7P-21

Reproduce it<br>Run any model against the games through the SDK or API.<br>API docs

Join us<br>Join our community DisCo, where you can track your progress on these puzzles and hang out with like-minded folk.

Citation<br>DiG-bench: Discovery in Games<br>Ruairidh M. Battleday, Kai Sandbrink, Jimi Cullen-Drohan, Zihan Yan, Timothy Muller, Clare Maguire, Ales Kubicek, Fraser Greenlee-Scott, Sukrit Sumant, Tri Dao, Jürgen Schmidhuber, Michal Valko, Joshua Tenenbaum, Thomas L. Griffiths, Zeb Kurth-Nelson, James C.R. Whittington<br>@misc {battleday2026dig,<br>title={DiG-bench: Discovery in Games},<br>author={Ruairidh M. Battleday and Kai Sandbrink and Jimi Cullen-Drohan and Zihan Yan and Timothy Muller and Clare Maguire and Ales Kubicek and Fraser Greenlee-Scott and Sukrit Sumant and Tri Dao and Jürgen Schmidhuber and Michal Valko and Joshua Tenenbaum and Thomas L. Griffiths and Zeb Kurth-Nelson and James C.R. Whittington},<br>year={2026},<br>eprint={2608.12593},<br>archivePrefix={arXiv},<br>primaryClass={cs.AI},<br>url={https://arxiv.org/abs/2608.12593},