Game Arena Go
We show this table for reference; we do not rank on it.
Kaggle Game Arena's Go board ranks models by head-to-head games of Go on a 13 by 13 board.
Evaluation results published by Kaggle. How Kaggle Benchmarks reruns tests.
Go rating on Game Arena Go v1 — Latest evaluation November 12, 2026
We mirror the published go rating view for Game Arena Go v1. Claude Fable 5.2 has the highest reported value at 1877. A rating gap smaller than the two rows' intervals is not a reliable difference. We do not use these results to rank models overall.
Claude Fable 5.2
−40/+49; Avg. Tokens/Turn 9999; Evaluated 2026-11-12
Gemini 4 Carbon
−43/+38; Avg. Tokens/Turn 36150; Evaluated 2026-11-12
GPT-6 Astra
−41/+42; Avg. Tokens/Turn 8702; Evaluated 2026-11-12
39 source rowsReasoningCurrentDisplay onlyUpdated Latest evaluation November 12, 2026
Go rating table (39 source rows)
| Rank | Model / configuration | Score | Parameters (B) | Open / closed |
|---|---|---|---|---|
| 1 | Claude Fable 5.2Creator not reported−40/+49; Avg. Tokens/Turn 9999; Evaluated 2026-11-12 | 1877 | Not reported | Unknown |
| 2 | Gemini 4 CarbonCreator not reported−43/+38; Avg. Tokens/Turn 36150; Evaluated 2026-11-12 | 1830 | Not reported | Unknown |
| 3 | GPT-6 AstraCreator not reported−41/+42; Avg. Tokens/Turn 8702; Evaluated 2026-11-12 | 1735 | Not reported | Unknown |
| 4 | Gemini 4 ArgonCreator not reported−48/+44; Avg. Tokens/Turn 11590; Evaluated 2026-11-12 | 1686 | Not reported | Unknown |
| 5 | Claude Fable 5.1Creator not reported−49/+42; Avg. Tokens/Turn 11706; Evaluated 2026-11-12 | 1601 | Not reported | Unknown |
| 6 | Claude Opus 5Creator not reported−55/+43; Avg. Tokens/Turn 17319; Evaluated 2026-11-12 | 1564 | Not reported | Unknown |
| 7 | Claude Opus 5.5Creator not reported−57/+43; Avg. Tokens/Turn 7144; Evaluated 2026-11-12 | 1543 | Not reported | Unknown |
| 8 | GPT-6.1 SolCreator not reported−55/+44; Avg. Tokens/Turn 8069; Evaluated 2026-11-12 | 1448 | Not reported | Unknown |
| 9 | Claude Sonnet 5.5Creator not reported−56/+47; Avg. Tokens/Turn 8904; Evaluated 2026-11-12 | 1442 | Not reported | Unknown |
| 10 | GPT-6.1 TerraCreator not reported−61/+44; Avg. Tokens/Turn 6698; Evaluated 2026-11-12 | 1202 | Not reported | Unknown |
| 11 | Gemini 3.8 FlashCreator not reported−63/+48; Avg. Tokens/Turn 23943; Evaluated 2026-11-12 | 1170 | Not reported | Unknown |
| 12 | Gemini 3.1 Pro PreviewCreator not reported−61/+47; Avg. Tokens/Turn 15921; Evaluated 2026-11-12 | 1147 | Not reported | Unknown |
| 13 | Claude Sonnet 5Creator not reported−63/+45; Avg. Tokens/Turn 28301; Evaluated 2026-11-12 | 1118 | Not reported | Unknown |
| 14 | Claude Opus 4.8Creator not reported−65/+45; Avg. Tokens/Turn 21003; Evaluated 2026-11-12 | 1116 | Not reported | Unknown |
| 15 | Claude Haiku 5.5Creator not reported−65/+40; Avg. Tokens/Turn 16865; Evaluated 2026-11-12 | 1076 | Not reported | Unknown |
| 16 | Grok 4.6Creator not reported−60/+41; Avg. Tokens/Turn 26931; Evaluated 2026-11-12 | 1023 | Not reported | Unknown |
| 17 | Gemini 3.7 FlashCreator not reported−61/+55; Avg. Tokens/Turn 11797; Evaluated 2026-11-12 | 1006 | Not reported | Unknown |
| 18 | GPT-5.6 SolCreator not reported−65/+44; Avg. Tokens/Turn 10505; Evaluated 2026-11-12 | 1005 | Not reported | Unknown |
| 19 | Grok 4.7Creator not reported−62/+49; Avg. Tokens/Turn 33141; Evaluated 2026-11-12 | 1004 | Not reported | Unknown |
| 20 | GPT-5.5Creator not reported−62/+46; Avg. Tokens/Turn 17749; Evaluated 2026-11-12 | 996 | Not reported | Unknown |
| 21 | Grok 4.5Creator not reported−61/+45; Avg. Tokens/Turn 23871; Evaluated 2026-11-12 | 904 | Not reported | Unknown |
| 22 | GPT-6 SolCreator not reported−62/+50; Avg. Tokens/Turn 8248; Evaluated 2026-11-12 | 900 | Not reported | Unknown |
| 23 | GPT-5.4Creator not reported−64/+49; Avg. Tokens/Turn 22438; Evaluated 2026-11-12 | 856 | Not reported | Unknown |
| 24 | Gemini 3.5 FlashCreator not reported−73/+61; Avg. Tokens/Turn 11700; Evaluated 2026-11-12 | 756 | Not reported | Unknown |
| 25 | Gemini 3.6 FlashCreator not reported−76/+56; Avg. Tokens/Turn 11414; Evaluated 2026-11-12 | 712 | Not reported | Unknown |
| 26 | Claude Opus 4.6Creator not reported−70/+53; Avg. Tokens/Turn 17586; Evaluated 2026-11-12 | 570 | Not reported | Unknown |
| 27 | GPT-5.2Creator not reported−70/+51; Avg. Tokens/Turn 18078; Evaluated 2026-11-12 | 561 | Not reported | Unknown |
| 28 | o3Creator not reported−74/+54; Avg. Tokens/Turn 22508; Evaluated 2026-11-12 | 555 | Not reported | Unknown |
| 29 | GPT-5.6 TerraCreator not reported−75/+53; Avg. Tokens/Turn 7557; Evaluated 2026-11-12 | 512 | Not reported | Unknown |
| 30 | Claude Opus 4.7Creator not reported−74/+52; Avg. Tokens/Turn 18134; Evaluated 2026-11-12 | 510 | Not reported | Unknown |
| 31 | Gemini 3 Flash PreviewCreator not reported−69/+59; Avg. Tokens/Turn 21590; Evaluated 2026-11-12 | 497 | Not reported | Unknown |
| 32 | GPT-6 LunaCreator not reported−75/+53; Avg. Tokens/Turn 8360; Evaluated 2026-11-12 | 474 | Not reported | Unknown |
| 33 | Grok 4.20 ReasoningCreator not reported−75/+55; Avg. Tokens/Turn 12578; Evaluated 2026-11-12 | 471 | Not reported | Unknown |
| 34 | GPT-5.6 LunaCreator not reported−88/+50; Avg. Tokens/Turn 11593; Evaluated 2026-11-12 | 446 | Not reported | Unknown |
| 35 | Claude Sonnet 4.5Creator not reported−84/+56; Avg. Tokens/Turn 10923; Evaluated 2026-11-12 | 325 | Not reported | Unknown |
| 36 | Claude Sonnet 4.6Creator not reported−84/+53; Avg. Tokens/Turn 14050; Evaluated 2026-11-12 | 296 | Not reported | Unknown |
| 37 | Grok 4.3Creator not reported−103/+72; Avg. Tokens/Turn 10245; Evaluated 2026-11-12 | 130 | Not reported | Unknown |
| 38 | GPT-5.4 miniCreator not reported−107/+75; Avg. Tokens/Turn 24323; Evaluated 2026-11-12 | 120 | Not reported | Unknown |
| 39 | GPT-5 miniCreator not reported−117/+81; Avg. Tokens/Turn 11560; Evaluated 2026-11-12 | 0 | Not reported | Unknown |
How to read this leaderboard
Use it to compare strategic planning and strict rule-following over a long game. The rating is relative to the models on this board; it is not a percentage and is not comparable with human Go ratings.
Operator receipt: 39 sourced rows are currently displayable on this page; the highest published score among these 39 source rows is Claude Fable 5.2 at 1877, which does not establish a market leader.
Honest limit: Ratings move as new games are played, and gaps smaller than the two rows' intervals are not reliable. The text harness, not a Go engine, decides how the board is shown, so results depend on it. One game is a narrow test; the pooled Game Arena board covers more games.
How to read Game Arena Go
Models play each other through OpenSpiel's Go environment with a text harness: each turn shows the game state, an ASCII board and the move history, and the model answers with one legal move or a pass. Games use area scoring with komi. The score is a Bradley-Terry rating fitted to game results, published with a bootstrapped interval that can be uneven above and below. Kaggle also publishes average tokens and cost per turn.
Ratings move as new games are played, and gaps smaller than the two rows' intervals are not reliable. The text harness, not a Go engine, decides how the board is shown, so results depend on it. One game is a narrow test; the pooled Game Arena board covers more games.
Snapshot
A rating gap smaller than the two rows' intervals is not a reliable difference. The table retains each source configuration and its reported value.
39 source rows are shown for Game Arena Go. The benchmark falls in the Reasoning category. We keep external benchmark mirrors separate from the weighted global scoring system, so these results remain source-specific evidence. Game Arena Go is currently displayed for reference but excluded from the scoring formula, so it does not directly affect overall rankings.
About Game Arena Go
Year
2026
Tasks
Head-to-head Go games on a 13 x 13 board
Format
Bradley-Terry rating from game outcomes
Difficulty
Strategic planning and rule-following
Models play each other through OpenSpiel's Go environment with a text harness: each turn shows the game state, an ASCII board and the move history, and the model answers with one legal move or a pass. Games use area scoring with komi. The score is a Bradley-Terry rating fitted to game results, published with a bootstrapped interval that can be uneven above and below. Kaggle also publishes average tokens and cost per turn. These results are display-only and do not enter overall or category rankings.
Freshness and provenance
Version
Game Arena Go 2026
Refresh cadence
Quarterly
Staleness state
Current
Question availability
Public benchmark set
BenchLM uses freshness metadata to decide whether a benchmark should still be treated as a strong differentiator, a benchmark to watch, or a display-only reference. For the full scoring policy, see the BenchLM methodology page.
Questions
What does Game Arena Go measure?
Kaggle Game Arena's Go board ranks models by head-to-head games of Go on a 13 by 13 board.
Which source row has the highest reported Game Arena Go value?
Claude Fable 5.2 has the highest reported Game Arena Go value at 1877. A rating gap smaller than the two rows' intervals is not a reliable difference. BenchLM shows this benchmark for display only and does not use it in overall rankings.
How many models are evaluated on Game Arena Go?
The Latest evaluation November 12, 2026 snapshot contains 39 AI models.
Know when it’s worth switching models
The model to choose, the cheaper alternative, and the release we would wait on.
Read a sample issueJoin 7,000+ subscribers.
One email each week. Unsubscribe anytime.