GameBench 2 — Current standings
AI model rankings
AI-written players, ranked by match results. Reasoning Variants ranks each model and reasoning setting separately.
- Updated
- 2026-10-26
- Model families
- 97
- Reasoning settings
- 261
- Player programs
- 2,053
- Public games
- 8
- Rated matches
- 525,020
Filtered view
Filters change this view; published ranks remain in the Base column.
Rankings
Reasoning settings
Detailed leaderboard
How to read
We ask each AI model to write a player program for every game. DuelLab runs those programs against each other. Higher scores mean stronger results against a fixed reference field of player programs.
Score is the 0–100 result across public games: expected points against the fixed reference field. Working-program strength averages only games where the model produced a working player program, leaving failed generation attempts out. Codegen shows the share that worked first try / worked after repair or resend / failed. Players counts the player programs behind a row, while Best is its highest single-game score. W / D / L shows rated wins, draws, and losses. More matches with existing programs do not add independent generation evidence. Where shown, the reasoning spread tells how much results vary across settings. Opponents played, hidden until you turn it on under Columns, shows how far the row's player programs are through the full round-robin of their games.
Partial 3/3 is an official Overall rank whose evidence is incomplete in another way, so compare it with care. Models with results for fewer than three reasoning settings get no Overall rank. Provisional rows do not receive an official rank. Capped draws reached the shared move cap and still count as draws. Filters can change the visible order, but the Official or Base column keeps the published rank. Cost / player is an estimate; select a value for its source and pricing details.
Columns
Choose the optional columns shown in this table.
Reasoning settings
Not yet ranked (31)
These entries have scores so far but do not yet meet the rules for an official rank, so they are kept out of the leaderboard above.
- 4Not every game attempted Some games have no player-program attempt: their generation is still running or stopped before them. The score so far covers only the games attempted.
- 16Some games cannot be scored A game's only generation attempts failed because of our test harness or the provider, not the model, and cannot currently be repeated. The score averages the other games.
- 2Every program failed None of the setting's generation attempts produced a working player program, so it scores 0.
- 8Too many programs failed Too few of its player programs compiled and ran to qualify for a rank.
- 1Matches still running Not enough matches have been played yet for a stable score.
Show all 31 entriesHide entries
| Model | Reasoning | Score so far | Why not ranked yet |
|---|---|---|---|
 Step 5 Preview | Medium | 83.0 | Not every game attempted |
 Kimi K3.1 Pro | High | 82.3 | Matches still running |
 Claude Haiku 5.5 | XHigh | 74.9 | Some games cannot be scored |
 Muse Spark Open | XHigh | 63.6 | Some games cannot be scored |
 Mercury 3 | High | 59.6 | Some games cannot be scored |
 MiMo-V2.5-Pro | Enabled | 46.8 | Some games cannot be scored |
| Mercury 3 | Medium | 45.0 | Some games cannot be scored |
| Kimi K2.7 Code | Enabled | 40.6 | Some games cannot be scored |
 Nex N2 Pro | None | 37.0 | Some games cannot be scored |
| MiMo-V2.5 | High | 35.9 | Some games cannot be scored |
| Step 5 Preview | Low | 34.3 | Not every game attempted |
| MiMo-V2.5 | Reasoning | 34.1 | Some games cannot be scored |
 Ling 3.1 Flash | None | 33.3 | Not every game attempted |
| Ling 3.0 Flash | Enabled | 30.6 | Some games cannot be scored |
| Ling 3.0 Flash | None | 30.2 | Not every game attempted |
| MiMo-V2.5 | Medium | 29.4 | Some games cannot be scored |
 Nemotron 3 Ultra 550B A55B | Medium | 27.7 | Some games cannot be scored |
| MiMo-V2.5 | Reasoning | 27.0 | Some games cannot be scored |
| Nex N2.5 Pro | High | 20.6 | Some games cannot be scored |
| Nex N2.5 Pro | None | 18.6 | Too many programs failed |
 Qwen3.8 27B | None | 17.5 | Too many programs failed |
| Nex N2 Pro | Enabled | 15.9 | Some games cannot be scored |
 Solar Mini 4 | XHigh | 13.8 | Too many programs failed |
 GPT-4.1 Nano | Reasoning | 11.8 | Too many programs failed |
| Nemotron 3.5 Super Vl | None | 11.7 | Too many programs failed |
 North Mini Code | None | 6.7 | Too many programs failed |
| Solar Mini 4 | Medium | 5.7 | Some games cannot be scored |
| Solar Mini 4 | None | 5.1 | Too many programs failed |
 Granite 4.2 8B | Low | 4.9 | Too many programs failed |
| Granite 4.2 8B | Non-thinking | 0.0 | Every program failed |
| Granite 4.2 8B | Full | 0.0 | Every program failed |
Scoring method
How this is scored
Each model is asked to create a program that can play every benchmark game.
The generated programs compete head-to-head, with both players receiving comparable opportunities.
Wins, draws, and losses produce a B1 (Bradley–Terry) rating for each program, which becomes a 0–100 score for each game: 100 × its expected points against the fixed reference field.
The main score is the mean across games. Every generation attempt counts: a game combines its working programs' scores with a 0 for each attempt whose player program could not be generated, and a reasoning setting where every attempt failed scores 0. A program rejected only for being too slow, or lost to a fault of the test machine, does not play and does not count. The share of working programs is shown on its own (Codegen).
Each model and reasoning setting keeps its own row in the standings.
Detailed tables provide match records, match counts, and other context for careful comparison.