Name: DuelLab GameBench 2 public release gb2-20260824-e05a1ff89a26
Creator: DuelLab
License: https://creativecommons.org/licenses/by/4.0/

Game profile

Xiangqi

Xiangqi is a 9x10 chess-family game with palaces, river restrictions, cannon screens, horse-leg and elephant-eye blocking, and flying-general legality.

Scores here compare player programs using DuelLab's encoded rules for Xiangqi; they are not a broad measure of human play strength.

By model and setting

GB2-04: Xiangqi by model and setting

Player programs are ranked by relative per-game score (0–100). Raw rating is shown as an advanced metric, alongside wins, draws, losses, and a per-game uncertainty index.

Reasoning level: Intensive Game: GB2-04: Xiangqi

GB2-04: Xiangqi - Intensive filter
Rank	Model	Score	Contributing programs	Raw Elo	W / D / L	Uncertainty
1	Claude Fable 5	100.0	1	1867.4	68 / 13 / 5	9.9
2	GPT-5.6 Sol	91.3	1	1813.5	68 / 8 / 2	12.5
3	Claude Opus 5	82.0	2	1756.1	102 / 21 / 17	15.6
4	Kimi K3	81.0	1	1734.8	82 / 47 / 15	0.0
5	Claude Opus 4.8	62.0	1	1623.8	35 / 29 / 16	11.8
6	Gemini 3.6 Flash	59.7	2	1597.9	129 / 74 / 83	0.0
7	GPT-5.4	59.3	1	1605.8	35 / 43 / 4	11.1
8	Claude Sonnet 5	58.9	1	1605.4	26 / 40 / 8	14.0
9	GPT-5.6 Terra	52.1	1	1559.7	27 / 40 / 13	11.8
10	Qwen3.8 Max	50.9	1	1595.3	9 / 10 / 1	65.4
11	Inkling Small	44.9	1	1563.3	7 / 9 / 1	74.1
12	Ox Alpha	42.0	3	1547.1	22 / 17 / 9	77.5
13	Inkling	41.1	1	1480.4	27 / 81 / 18	0.6
14	MiMo-V2.5	40.8	1	1494.5	6 / 43 / 9	21.3
15	Gemini 3.1 Flash Lite	40.4	1	1484.8	6 / 54 / 18	12.5
16	DeepSeek V4 Pro	39.5	1	1479.6	9 / 49 / 18	13.2
17	GPT-OSS 120B	38.8	1	1478.3	14 / 32 / 20	17.3
18	GPT 5	38.7	1	1474.3	7 / 52 / 17	13.2
19	Gemini 3.5 Flash Lite	36.7	2	1451.4	34 / 163 / 59	0.3
20	Nemotron 3 Ultra 550B A55B	35.8	1	1455.2	6 / 56 / 16	12.5
21	Claude Opus 4.5	35.6	1	1441.4	17 / 79 / 55	0.0
22	Gemini 3.5 Flash	34.8	1	1447.5	21 / 29 / 32	11.1
23	Grok 4.5	32.4	1	1421.5	35 / 59 / 52	0.0
24	Hy3 Preview	31.9	1	1431.3	7 / 49 / 18	14.0
25	GPT-5.5	31.3	1	1426.8	6 / 43 / 27	13.2
26	GPT-5.6 Luna	30.6	1	1421.7	11 / 44 / 23	12.5

Underlying results

Individual player programs

Each row is one accepted player program for GB2-04: Xiangqi. A model can appear more than once when it produced more than one program. Score is the normalized 0–100 result; Elo and W / D / L show the underlying play record. Confidence and evidence can be blank when too few compatible matches are available. Draws at the move limit count; one-sided model faults are listed separately.

Search model Status Sort Showing 32 player programs.

Player ID	Model	Status	Score	Elo	Confidence	W / D / L	Rated / attempted	Opponents	Faults / players	Generation
E-0ca15b39b465b370	Claude Fable 5	Rated	100.0	1867	—	68 / 13 / 5	86 / 86	37	0 / 1	First try · 2026-07-03
E-e3449ff084425b16	Claude Opus 4.5	Waiting	—	—	—	0 / 0 / 0	0 / 0	0	0 / 1	First try · 2026-07-02
E-f40463f635c43598	Claude Opus 4.5	Rated	35.6	1441	—	17 / 79 / 55	151 / 151	56	0 / 1	First try · 2026-07-21
E-6344242dae68be1b	Claude Opus 4.8	Rated	62.0	1624	—	35 / 29 / 16	80 / 80	36	0 / 1	First try · 2026-07-03
E-4c98514bdb0eaf10	Claude Opus 5	Rated	73.4	1701	—	48 / 9 / 11	68 / 68	30	0 / 1	First try · 2026-07-25
E-a02883202d334eec	Claude Opus 5	Rated	90.7	1811	—	54 / 12 / 6	72 / 72	30	0 / 1	First try · 2026-07-25
E-74d5848169176ddd	Claude Sonnet 5	Rated	58.9	1605	—	26 / 40 / 8	74 / 74	36	0 / 1	First try · 2026-08-09
E-38bc194a470fbf62	DeepSeek V4 Pro	Rated	39.5	1480	—	9 / 49 / 18	76 / 76	38	0 / 1	Repaired · 2026-07-02
E-2be6f5c895691cb3	Gemini 3.1 Flash Lite	Rated	40.4	1485	—	6 / 54 / 18	78 / 78	35	0 / 1	Repaired · 2026-07-02
E-8e95359e4cbd1851	Gemini 3.5 Flash	Rated	34.8	1447	—	21 / 29 / 32	82 / 82	38	0 / 1	Repaired · 2026-07-02
E-39ebef34ae0ea1cb	Gemini 3.5 Flash Lite	Rated	35.9	1446	—	10 / 89 / 29	128 / 129	49	0 / 1	First try · 2026-07-24
E-953992515ec1424d	Gemini 3.5 Flash Lite	Rated	37.5	1457	—	24 / 74 / 30	128 / 128	50	0 / 1	First try · 2026-07-24
E-a5a73c81dfcf4540	Gemini 3.6 Flash	Rated	42.0	1484	—	44 / 35 / 59	138 / 138	51	0 / 1	First try · 2026-07-24
E-ccfb6c993d1622b0	Gemini 3.6 Flash	Rated	77.4	1711	—	85 / 39 / 24	148 / 151	56	0 / 1	Repaired · 2026-07-24
E-7e3da77777a95e31	GPT 5	Rated	38.7	1474	—	7 / 52 / 17	76 / 76	37	0 / 1	Repaired · 2026-07-02
E-8d6c86dd58efc4be	GPT-5.4	Rated	59.3	1606	—	35 / 43 / 4	82 / 85	37	0 / 1	First try · 2026-08-09
E-64bf5daf186f8119	GPT-5.5	Rated	31.3	1427	—	6 / 43 / 27	76 / 76	38	0 / 1	First try · 2026-07-02
E-0a9286b0c21ae9be	GPT-5.6 Luna	Rated	30.6	1422	—	11 / 44 / 23	78 / 78	36	0 / 1	First try · 2026-08-09
E-6b542c5d0dcf2279	GPT-5.6 Sol	Rated	91.3	1814	—	68 / 8 / 2	78 / 78	36	0 / 1	First try · 2026-08-09
E-b85c80e7475ef9ec	GPT-5.6 Terra	Rated	52.1	1560	—	27 / 40 / 13	80 / 80	38	0 / 1	First try · 2026-07-10
E-601725df2c678708	Kimi K3	Rated	81.0	1735	—	82 / 47 / 15	144 / 144	54	0 / 1	Repaired · 2026-07-24
E-958dea26b779a042	Nemotron 3 Ultra 550B A55B	Rated	35.8	1455	—	6 / 56 / 16	78 / 78	38	0 / 1	Repaired · 2026-07-11
E-3a27113d21c73b01	GPT-OSS 120B	Rated	38.8	1478	—	14 / 32 / 20	66 / 66	32	0 / 1	Repaired · 2026-07-02
E-2c5709bd26aef5e5	Qwen3.8 Max	Limited data	50.9	1595	—	9 / 10 / 1	20 / 20	11	0 / 1	First try · 2026-08-18
E-454f79b70ee12fe0	Ox Alpha	Rated	59.6	1661	—	14 / 2 / 0	16 / 16	8	0 / 1	First try · 2026-08-22
E-6a2b79a57d322dae	Ox Alpha	Rated	32.2	1484	—	4 / 7 / 5	16 / 16	8	0 / 1	First try · 2026-08-22
E-d89754cdf98b2ae2	Ox Alpha	Rated	34.1	1496	—	4 / 8 / 4	16 / 16	8	0 / 1	First try · 2026-08-22
E-51c104b34ef2a244	Hy3 Preview	Rated	31.9	1431	—	7 / 49 / 18	74 / 74	36	0 / 1	First try · 2026-07-02
E-f130161523da1e6b	Inkling	Rated	41.1	1480	—	27 / 81 / 18	126 / 126	49	0 / 1	First try · 2026-07-24
E-df6d76539ff35338	Inkling Small	Limited data	44.9	1563	—	7 / 9 / 1	17 / 17	8	0 / 1	Repaired · 2026-08-18
E-07caaa12175b0b3c	Grok 4.5	Rated	32.4	1422	—	35 / 59 / 52	146 / 146	55	0 / 1	First try · 2026-07-24
E-f8929b12716a2221	MiMo-V2.5	Rated	40.8	1495	—	6 / 43 / 9	58 / 58	29	0 / 1	Repaired · 2026-07-02