By model and setting

Each model and reasoning setting appears once. If a model produced more than one player program for this game, the row stands for all of them: Programs counts the working attempts out of the counted attempts, Score and Raw rating average the programs (weighted by attempts; the small range is the lowest and highest program rating), and the combined record adds up their matches. Score is relative within this game (0–100). Scores here are working-program strength: failed generation attempts are left out and count as 0 only in a model's headline score.

GB2-04: Xiangqi - Reasoning Variants
Rank Model Reasoning Score Programs Raw rating Combined record (W / D / L)
1Claude Fable 5.2XHigh99.81 working of 1 attempt2626.8578 / 0 / 0
2Claude Fable 5.2Medium95.61 working of 1 attempt2089.5532 / 33 / 13
3Gemini 4 HydrogenXHigh95.61 working of 1 attempt2089.5530 / 37 / 11
4Claude Opus 5.5XHigh94.61 working of 1 attempt2048.4516 / 52 / 10
5GPT-6.1 SolXHigh94.21 working of 1 attempt2034.2518 / 43 / 17
6Grok 4.7Medium92.21 working of 1 attempt1971.1497 / 59 / 22
7GPT-6 AstraXHigh90.71 working of 1 attempt1934.8494 / 47 / 37
8GPT-6 AstraMedium90.31 working of 1 attempt1923.6486 / 57 / 35
9Claude Fable 5.2Low89.81 working of 1 attempt1912.9478 / 67 / 33
10Claude Fable 5.1XHigh88.51 working of 1 attempt1886.1473 / 61 / 44
11GPT-6 SolXHigh87.21 working of 1 attempt1861.4466 / 59 / 53
12GPT-6.1 SolMedium86.61 working of 1 attempt1849.8466 / 51 / 61
13GPT-6.1 Terra ProNone85.51 working of 1 attempt1831.8445 / 80 / 53
14Claude Opus 5.5Medium85.31 working of 1 attempt1827.7456 / 55 / 67
15Pareto 26.10 PreviewDefault85.01 working of 1 attempt1822.4454 / 55 / 69
16Pareto 26.10Default84.91 working of 1 attempt1821.1443 / 76 / 59
17GPT-6.1 Terra ProXHigh84.21 working of 1 attempt1810.8443 / 68 / 67
18Claude Fable 5XHigh83.81 working of 1 attempt1804.5432 / 85 / 61
19GPT-6.1 Terra ProMedium83.71 working of 1 attempt1803.3435 / 78 / 65
20GPT-6.1 TerraXHigh83.11 working of 1 attempt1793.5442 / 56 / 80
21GPT-6.1 TerraNone83.01 working of 1 attempt1792.3439 / 61 / 78
22Qwen3.8 Max 0902XHigh82.71 working of 1 attempt1787.5417 / 101 / 60
23GPT-6 AstraLow82.61 working of 1 attempt1786.3419 / 96 / 63
24GLM-5.3High82.51 working of 1 attempt1785.1412 / 109 / 57
25Claude Sonnet 5.5XHigh82.41 working of 1 attempt1784.0412 / 108 / 58

Individual player programs

Each row is one accepted player program for GB2-04: Xiangqi; a model can have more than one. Score is the 0–100 result in this game, and its rating and W / D / L record show how it played. Scores are working-program strength: failed generation attempts have no row and count only in a model's headline score.

288 player programs.

ModelReasoningStatusScoreRatingW / D / LMatchesOpponentsNotesGenerated
Claude Fable 5.2E-8d188ef917c7254fXHighRated Interval validated99.82627578 / 0 / 0578 / 578289—2026-11-09First try
Claude Fable 5.2E-4a4845a6d789cc84MediumRated Interval validated95.62089532 / 33 / 13578 / 578289—2026-11-09First try
Gemini 4 HydrogenE-c22376666478108aXHighRated Interval validated95.62089530 / 37 / 11578 / 578289—2026-11-12First try
Claude Opus 5.5E-d512f5c88a2f7092XHighRated Interval validated94.62048516 / 52 / 10578 / 578289—2026-09-22Repaired
GPT-6.1 SolE-0beb3abad220fddbXHighRated Interval validated94.22034518 / 43 / 17578 / 578289—2026-10-01First try
Grok 4.7E-3fd10aaafda5a40dMediumRated Interval validated92.21971497 / 59 / 22578 / 578289—2026-09-30First try
GPT-6 AstraE-154ed79df2f33a60XHighRated Interval validated90.71935494 / 47 / 37578 / 578289—2026-09-05First try
GPT-6 AstraE-8863403f24a275fdMediumRated Interval validated90.31924486 / 57 / 35578 / 578289—2026-09-05First try
Claude Fable 5.2E-9f3e25b5487e28f2LowRated Interval validated89.81913478 / 67 / 33578 / 5782891 capped draw2026-11-09First try
Claude Fable 5.1E-6c73f19c641d4021XHighRated Interval validated88.51886473 / 61 / 44578 / 578289—2026-09-08First try
GPT-6 SolE-ca73d8cf5d48a8e6XHighRated Interval validated87.21861466 / 59 / 53578 / 578289—2026-09-22First try
Ox AlphaE-454f79b70ee12fe0MaxRated Interval validated87.01857440 / 108 / 30578 / 578289—2026-08-22First try
GPT-6.1 SolE-ccfb3a267deaea31MediumRated Interval validated86.61850466 / 51 / 61578 / 578289—2026-10-01First try
Claude Fable 5.1E-7002caf3e34510f6XHighRated Interval validated85.81836463 / 47 / 68578 / 578289—2026-09-03First try
Claude Opus 5E-a02883202d334eecXHighRated Interval validated85.51832431 / 108 / 39578 / 5782891 capped draw2026-07-25First try
GPT-6.1 Terra ProE-a6e02e1d90155ce9NoneRated Interval validated85.51832445 / 80 / 53578 / 578289—2026-11-01First try
Claude Opus 5.5E-d01a435c350a4a98MediumRated Interval validated85.31828456 / 55 / 67578 / 578289—2026-09-22First try
Pareto 26.10 PreviewE-4d1e4650d78a4cb3DefaultRated Interval validated85.01822454 / 55 / 69578 / 578289—2026-10-09First try
Pareto 26.10E-334062be20fe1540DefaultRated Interval validated84.91821443 / 76 / 59578 / 578289—2026-10-21First try
GPT-6.1 Terra ProE-d1b340cf74b6f8b1XHighRated Interval validated84.21811443 / 68 / 67578 / 578289—2026-11-01First try
Claude Fable 5E-e34f220d213bb5a8XHighRated Interval validated83.81805432 / 85 / 61578 / 578289—2026-09-03First try
GPT-6.1 Terra ProE-3099c50d52790547MediumRated Interval validated83.71803435 / 78 / 65578 / 578289—2026-11-01First try
GPT-6.1 TerraE-4a072d3fd4ee3bdcXHighRated Interval validated83.11793442 / 56 / 80578 / 578289—2026-11-01First try
GPT-6.1 TerraE-adf3dba4080d651cNoneRated Interval validated83.01792439 / 61 / 78578 / 578289—2026-11-01First try
Qwen3.8 Max 0902E-924ba680b1c3e7cdXHighRated Interval validated82.71788417 / 101 / 60578 / 578289—2026-10-09First try