Reasoning benchmark / Epoch AI

Mystery Game Puzzles

Lumina Verified

Single best move across 100 undisclosed game puzzles. Published evaluations by Epoch AI.

Version Puzzles-100Verified 31 October 2026142 published resultsOfficial source

Mystery Game Puzzles · Lumina Score

Epoch AI · 100 game positions

LuminaBench

20 of 142 configurations

Epoch AI mean accuracy and published standard error; unspecified effort remains unknown. Missing metrics are omitted from their graph; quality scores remain available.

Results table

Mystery Game Puzzles Puzzles-100. Original publisher values and provisional Lumina conversion.
Rank
1	GPT-6 AstraMax	1,273	84%	±3.68 pp SE
2	GPT-6.1 SolMax	1,255	80%	±4.02 pp SE
3	GPT-6.1 TerraMax	1,228	74%	±4.41 pp SE
4	Claude Opus 5.5Max	1,215	71%	±4.56 pp SE
5	Claude Sonnet 5.5Max	1,188	65%	±4.79 pp SE
6	Gemini 4 ArgonHigh	1,183	64%	±4.82 pp SE
7	Claude Opus 5Max	1,161	59%	±4.94 pp SE
8	GPT-5.6 SolMax	1,156	58%	±4.96 pp SE
8	Claude Fable 5.1Max	1,156	58%	±4.96 pp SE
10	GPT-6 SolMax	1,147	56%	±4.99 pp SE
10	GPT-5.5xHigh	1,147	56%	±4.99 pp SE
12	GPT-5.5High	1,129	52%	±5.02 pp SE
12	Claude Fable 5Max	1,129	52%	±5.02 pp SE
14	Muse WonderMax	1,125	51%	±5.02 pp SE
15	DeepSeek V4.1 Flash 1017Max	1,116	49%	±5.02 pp SE
16	Gemini 3.8 FlashHigh	1,107	47%	±5.02 pp SE
17	DeepSeek V4 Pro 0813Max	1,089	43%	±4.98 pp SE
17	DeepSeek V4.1 FlashMax	1,089	43%	±4.98 pp SE
19	qwen3.8 MaxxHigh	1,066	38%	±4.88 pp SE
20	Claude Opus 5Not specified	1,062	37%	±4.85 pp SE
20	GPT-5.4xHigh	1,062	37%	±4.85 pp SE
20	Gemini 3.7 FlashHigh	1,062	37%	±4.85 pp SE
23	Claude Opus 4.8Max	1,057	36%	±4.82 pp SE
24	GPT-5.6 TerraMax	1,053	35%	±4.79 pp SE
24	Claude Sonnet 5Max	1,053	35%	±4.79 pp SE
26	DeepSeek V4 Flash 0731Max	1,048	34%	±4.76 pp SE
26	Grok 4.6xHigh	1,048	34%	±4.76 pp SE
26	Gemini 3.1 Pro (Preview)High	1,048	34%	±4.76 pp SE
29	GPT-5.6 SolNone	1,044	33%	±4.73 pp SE
29	GLM 5.3Max	1,044	33%	±4.73 pp SE
31	qwen3.7 MaxNot specified	1,039	32%	±4.69 pp SE
31	Gemini 3.5 FlashHigh	1,039	32%	±4.69 pp SE
31	Gemini 3.1 Pro (Preview)Medium	1,039	32%	±4.69 pp SE
34	Claude Opus 4.8xHigh	1,035	31%	±4.65 pp SE
35	Claude Haiku 5.5Max	1,030	30%	±4.61 pp SE
35	Gemini 3.6 FlashHigh	1,030	30%	±4.61 pp SE
37	Gemini 3.1 Pro (Preview)Low	1,026	29%	±4.56 pp SE
37	o3High	1,026	29%	±4.56 pp SE
37	Grok 4.7xHigh	1,026	29%	±4.56 pp SE
40	GPT-5.4Medium	1,021	28%	±4.51 pp SE
40	Gemini 3.5 FlashLow	1,021	28%	±4.51 pp SE
40	GPT-5.5Low	1,021	28%	±4.51 pp SE
40	Claude 4.7Max	1,021	28%	±4.51 pp SE
44	Kimi K3Max	1,012	26%	±4.41 pp SE
44	qwen3.7 MaxNone	1,012	26%	±4.41 pp SE
44	GPT-5.6 SolLow	1,012	26%	±4.41 pp SE
44	Gemini 3 Flash PreviewLow	1,012	26%	±4.41 pp SE
48	Gemini 3.6 FlashMinimal	1,008	25%	±4.35 pp SE
48	Claude Opus 4.6Max	1,008	25%	±4.35 pp SE
48	Muse Spark 1.3Max	1,008	25%	±4.46 pp SE
48	Gemini 3 Flash PreviewMinimal	1,008	25%	±4.35 pp SE
52	GPT-5.2High	999	23%	±4.23 pp SE
52	o3Medium	999	23%	±4.23 pp SE
52	Gemini 3.6 FlashLow	999	23%	±4.23 pp SE
52	GPT-5High	999	23%	±4.23 pp SE
56	Claude Opus 4.548k	994	22%	±4.16 pp SE
56	GPT-5.2Medium	994	22%	±4.16 pp SE
56	qwen3.6 35b a3bNone	994	22%	±4.16 pp SE
59	Claude Opus 4 1 2025080524k	990	21%	±4.09 pp SE
59	GPT-5.6 LunaMax	990	21%	±4.09 pp SE
61	qwen3.5 FlashNot specified	985	20%	±4.02 pp SE
61	Gemini 3 Flash PreviewHigh	985	20%	±4.02 pp SE
61	GPT-5.6 LunaNone	985	20%	±4.02 pp SE
61	Nemotron 3 UltraNot specified	985	20%	±4.02 pp SE
65	GLM-5.2Low	981	19%	±3.94 pp SE
65	GPT-5.6 TerraLow	981	19%	±3.94 pp SE
65	GPT-5.1Low	981	19%	±3.94 pp SE
65	qwen3.6 Max PreviewNone	981	19%	±3.94 pp SE
65	Gemini 3.5 Flash-LiteLow	981	19%	±3.94 pp SE
70	qwen3.6 FlashNone	977	18%	±3.86 pp SE
70	GLM-5.2Minimal	977	18%	±3.86 pp SE
70	GLM-5.2None	977	18%	±3.86 pp SE
70	qwen3.5 397b a17bNone	977	18%	±3.86 pp SE
70	GPT-5.5None	977	18%	±3.86 pp SE
70	Kimi k2.6Not specified	977	18%	±3.86 pp SE
76	Claude Sonnet 4.548k	972	17%	±3.78 pp SE
76	qwen3.7 PlusNone	972	17%	±3.78 pp SE
76	DeepSeek V4 ProNone	972	17%	±3.78 pp SE
76	qwen3.7 PlusNot specified	972	17%	±3.78 pp SE
76	GPT-5.6 LunaLow	972	17%	±3.78 pp SE
76	GPT-5.4Low	972	17%	±3.78 pp SE
76	Mistral Large 4 0High	972	17%	±3.78 pp SE
76	Qwen3.5-122B-A10BNone	972	17%	±3.78 pp SE
76	qwen3.5 PlusNone	972	17%	±3.78 pp SE
85	GPT-5Low	968	16%	±3.68 pp SE
85	GPT-5.1Medium	968	16%	±3.68 pp SE
85	Claude Sonnet 5Not specified	968	16%	±3.68 pp SE
85	qwen3.5 FlashNone	968	16%	±3.68 pp SE
85	GPT-5.4None	968	16%	±3.68 pp SE
85	qwen3.5 PlusNot specified	968	16%	±3.68 pp SE
85	Claude Sonnet 4.6Low	968	16%	±3.68 pp SE
92	GPT-5.1None	963	15%	±3.59 pp SE
92	GPT-5Medium	963	15%	±3.59 pp SE
92	qwen3.7 FlashNot specified	963	15%	±3.59 pp SE
92	Claude Opus 4.6Not specified	963	15%	±3.59 pp SE
92	GLM-5.2Medium	963	15%	±3.59 pp SE
97	Muse Spark 1.3xHigh	959	14.14%	±3.52 pp SE
97	GPT-5.6 TerraNone	959	14%	±3.49 pp SE
97	GPT-5.2None	959	14%	±3.49 pp SE
97	GPT-5Minimal	959	14%	±3.49 pp SE
97	qwen3.7 FlashNone	959	14%	±3.49 pp SE
97	Claude Sonnet 4.6Not specified	959	14%	±3.49 pp SE
103	Claude 4.7Not specified	954	13%	±3.38 pp SE
104	qwen3.6 FlashNot specified	950	12%	±3.27 pp SE
104	Kimi k2.6None	950	12%	±3.27 pp SE
104	GPT-5.6 LunaMedium	950	12%	±3.27 pp SE
104	Kimi K3Low	950	12%	±3.27 pp SE
104	GPT-4 0613Not specified	950	12%	±3.27 pp SE
104	Gemini 3.5 Flash-LiteMinimal	950	12%	±3.27 pp SE
104	qwen3.6 PlusNone	950	12%	±3.27 pp SE
104	GPT-4o-miniNot specified	950	12%	±3.27 pp SE
112	GPT-5.4 MiniNone	945	11%	±3.14 pp SE
113	GPT-5.2Low	941	10%	±3.02 pp SE
113	GPT-5 MiniMinimal	941	10%	±3.02 pp SE
113	GPT-5.6 TerraMedium	941	10%	±3.02 pp SE
116	Qwen3 235B A22B Thinking 2507Not specified	936	9%	±2.88 pp SE
116	GPT-5.4 NanoNone	936	9%	±2.88 pp SE
116	GPT-5 NanoLow	936	9%	±2.88 pp SE
116	GPT-5 NanoMedium	936	9%	±2.88 pp SE
120	GPT-5.4 MiniLow	932	8%	±2.73 pp SE
120	GPT-5 NanoHigh	932	8%	±2.73 pp SE
120	MiniMax M3None	932	8%	±2.73 pp SE
120	GLM 5.3 FlashMax	932	8%	±2.73 pp SE
124	GPT-5.4 MiniMedium	927	7%	±2.56 pp SE
124	GPT-6 LunaMax	927	7%	±2.56 pp SE
124	qwen3.6 27bNone	927	7%	±2.56 pp SE
124	MiniMax M3Not specified	927	7%	±2.56 pp SE
124	GPT-4.1-MiniNot specified	927	7%	±2.56 pp SE
124	Claude Opus 4.6Low	927	7%	±2.56 pp SE
124	o3-miniHigh	927	7%	±2.56 pp SE
131	GPT-5.4 NanoMedium	923	6%	±2.39 pp SE
131	Inkling SmallxHigh	923	6%	±2.39 pp SE
133	GPT-5.4 NanoHigh	918	5%	±2.19 pp SE
133	o4-miniHigh	918	5%	±2.19 pp SE
133	GPT-5 NanoMinimal	918	5%	±2.19 pp SE
133	qwen3 Max 2025 09 23Not specified	918	5%	±2.19 pp SE
133	GPT-5 MiniHigh	918	5%	±2.19 pp SE
138	GPT-5 MiniMedium	914	4%	±1.97 pp SE
139	GPT-5.4 NanoLow	909	3%	±1.71 pp SE
139	GPT-3.5 Turbo 0125Not specified	909	3%	±1.71 pp SE
141	GPT-Oss 120bMedium	905	2%	±1.41 pp SE
142	GPT-Oss 120bHigh	896	0%	±0 pp SE

1 · GPT-6 AstraMax1273

Accuracy: 84%
Standard error: ±3.68 pp SE
Evidence: Model profile ↗

2 · GPT-6.1 SolMax1255

Accuracy: 80%
Standard error: ±4.02 pp SE
Evidence: Model profile ↗

3 · GPT-6.1 TerraMax1228

Accuracy: 74%
Standard error: ±4.41 pp SE
Evidence: Model profile ↗

4 · Claude Opus 5.5Max1215

Accuracy: 71%
Standard error: ±4.56 pp SE
Evidence: Model profile ↗

5 · Claude Sonnet 5.5Max1188

Accuracy: 65%
Standard error: ±4.79 pp SE
Evidence: Model profile ↗

6 · Gemini 4 ArgonHigh1183

Accuracy: 64%
Standard error: ±4.82 pp SE
Evidence: Model profile ↗

7 · Claude Opus 5Max1161

Accuracy: 59%
Standard error: ±4.94 pp SE
Evidence: Model profile ↗

8 · GPT-5.6 SolMax1156

Accuracy: 58%
Standard error: ±4.96 pp SE
Evidence: Model profile ↗

8 · Claude Fable 5.1Max1156

Accuracy: 58%
Standard error: ±4.96 pp SE
Evidence: Model profile ↗

10 · GPT-6 SolMax1147

Accuracy: 56%
Standard error: ±4.99 pp SE
Evidence: Model profile ↗

10 · GPT-5.5xHigh1147

Accuracy: 56%
Standard error: ±4.99 pp SE
Evidence: Model profile ↗

12 · GPT-5.5High1129

Accuracy: 52%
Standard error: ±5.02 pp SE
Evidence: Model profile ↗

12 · Claude Fable 5Max1129

Accuracy: 52%
Standard error: ±5.02 pp SE
Evidence: Model profile ↗

14 · Muse WonderMax1125

Accuracy: 51%
Standard error: ±5.02 pp SE
Evidence: Model profile ↗

15 · DeepSeek V4.1 Flash 1017Max1116

Accuracy: 49%
Standard error: ±5.02 pp SE
Evidence: Model profile ↗

16 · Gemini 3.8 FlashHigh1107

Accuracy: 47%
Standard error: ±5.02 pp SE
Evidence: Model profile ↗

17 · DeepSeek V4 Pro 0813Max1089

Accuracy: 43%
Standard error: ±4.98 pp SE
Evidence: Model profile ↗

17 · DeepSeek V4.1 FlashMax1089

Accuracy: 43%
Standard error: ±4.98 pp SE
Evidence: Model profile ↗

19 · qwen3.8 MaxxHigh1066

Accuracy: 38%
Standard error: ±4.88 pp SE
Evidence: Model profile ↗

20 · Claude Opus 5Not specified1062

Accuracy: 37%
Standard error: ±4.85 pp SE
Evidence: Model profile ↗

20 · GPT-5.4xHigh1062

Accuracy: 37%
Standard error: ±4.85 pp SE
Evidence: Model profile ↗

20 · Gemini 3.7 FlashHigh1062

Accuracy: 37%
Standard error: ±4.85 pp SE
Evidence: Model profile ↗

23 · Claude Opus 4.8Max1057

Accuracy: 36%
Standard error: ±4.82 pp SE
Evidence: Model profile ↗

24 · GPT-5.6 TerraMax1053

Accuracy: 35%
Standard error: ±4.79 pp SE
Evidence: Model profile ↗

24 · Claude Sonnet 5Max1053

Accuracy: 35%
Standard error: ±4.79 pp SE
Evidence: Model profile ↗

26 · DeepSeek V4 Flash 0731Max1048

Accuracy: 34%
Standard error: ±4.76 pp SE
Evidence: Model profile ↗

26 · Grok 4.6xHigh1048

Accuracy: 34%
Standard error: ±4.76 pp SE
Evidence: Model profile ↗

26 · Gemini 3.1 Pro (Preview)High1048

Accuracy: 34%
Standard error: ±4.76 pp SE
Evidence: Model profile ↗

29 · GPT-5.6 SolNone1044

Accuracy: 33%
Standard error: ±4.73 pp SE
Evidence: Model profile ↗

29 · GLM 5.3Max1044

Accuracy: 33%
Standard error: ±4.73 pp SE
Evidence: Model profile ↗

31 · qwen3.7 MaxNot specified1039

Accuracy: 32%
Standard error: ±4.69 pp SE
Evidence: Model profile ↗

31 · Gemini 3.5 FlashHigh1039

Accuracy: 32%
Standard error: ±4.69 pp SE
Evidence: Model profile ↗

31 · Gemini 3.1 Pro (Preview)Medium1039

Accuracy: 32%
Standard error: ±4.69 pp SE
Evidence: Model profile ↗

34 · Claude Opus 4.8xHigh1035

Accuracy: 31%
Standard error: ±4.65 pp SE
Evidence: Model profile ↗

35 · Claude Haiku 5.5Max1030

Accuracy: 30%
Standard error: ±4.61 pp SE
Evidence: Model profile ↗

35 · Gemini 3.6 FlashHigh1030

Accuracy: 30%
Standard error: ±4.61 pp SE
Evidence: Model profile ↗

37 · Gemini 3.1 Pro (Preview)Low1026

Accuracy: 29%
Standard error: ±4.56 pp SE
Evidence: Model profile ↗

37 · o3High1026

Accuracy: 29%
Standard error: ±4.56 pp SE
Evidence: Model profile ↗

37 · Grok 4.7xHigh1026

Accuracy: 29%
Standard error: ±4.56 pp SE
Evidence: Model profile ↗

40 · GPT-5.4Medium1021

Accuracy: 28%
Standard error: ±4.51 pp SE
Evidence: Model profile ↗

40 · Gemini 3.5 FlashLow1021

Accuracy: 28%
Standard error: ±4.51 pp SE
Evidence: Model profile ↗

40 · GPT-5.5Low1021

Accuracy: 28%
Standard error: ±4.51 pp SE
Evidence: Model profile ↗

40 · Claude 4.7Max1021

Accuracy: 28%
Standard error: ±4.51 pp SE
Evidence: Model profile ↗

44 · Kimi K3Max1012

Accuracy: 26%
Standard error: ±4.41 pp SE
Evidence: Model profile ↗

44 · qwen3.7 MaxNone1012

Accuracy: 26%
Standard error: ±4.41 pp SE
Evidence: Model profile ↗

44 · GPT-5.6 SolLow1012

Accuracy: 26%
Standard error: ±4.41 pp SE
Evidence: Model profile ↗

44 · Gemini 3 Flash PreviewLow1012

Accuracy: 26%
Standard error: ±4.41 pp SE
Evidence: Model profile ↗

48 · Gemini 3.6 FlashMinimal1008

Accuracy: 25%
Standard error: ±4.35 pp SE
Evidence: Model profile ↗

48 · Claude Opus 4.6Max1008

Accuracy: 25%
Standard error: ±4.35 pp SE
Evidence: Model profile ↗

48 · Muse Spark 1.3Max1008

Accuracy: 25%
Standard error: ±4.46 pp SE
Evidence: Model profile ↗

48 · Gemini 3 Flash PreviewMinimal1008

Accuracy: 25%
Standard error: ±4.35 pp SE
Evidence: Model profile ↗

52 · GPT-5.2High999

Accuracy: 23%
Standard error: ±4.23 pp SE
Evidence: Model profile ↗

52 · o3Medium999

Accuracy: 23%
Standard error: ±4.23 pp SE
Evidence: Model profile ↗

52 · Gemini 3.6 FlashLow999

Accuracy: 23%
Standard error: ±4.23 pp SE
Evidence: Model profile ↗

52 · GPT-5High999

Accuracy: 23%
Standard error: ±4.23 pp SE
Evidence: Model profile ↗

56 · Claude Opus 4.548k994

Accuracy: 22%
Standard error: ±4.16 pp SE
Evidence: Model profile ↗

56 · GPT-5.2Medium994

Accuracy: 22%
Standard error: ±4.16 pp SE
Evidence: Model profile ↗

56 · qwen3.6 35b a3bNone994

Accuracy: 22%
Standard error: ±4.16 pp SE
Evidence: Model profile ↗

59 · Claude Opus 4 1 2025080524k990

Accuracy: 21%
Standard error: ±4.09 pp SE
Evidence: Model profile ↗

59 · GPT-5.6 LunaMax990

Accuracy: 21%
Standard error: ±4.09 pp SE
Evidence: Model profile ↗

61 · qwen3.5 FlashNot specified985

Accuracy: 20%
Standard error: ±4.02 pp SE
Evidence: Model profile ↗

61 · Gemini 3 Flash PreviewHigh985

Accuracy: 20%
Standard error: ±4.02 pp SE
Evidence: Model profile ↗

61 · GPT-5.6 LunaNone985

Accuracy: 20%
Standard error: ±4.02 pp SE
Evidence: Model profile ↗

61 · Nemotron 3 UltraNot specified985

Accuracy: 20%
Standard error: ±4.02 pp SE
Evidence: Model profile ↗

65 · GLM-5.2Low981

Accuracy: 19%
Standard error: ±3.94 pp SE
Evidence: Model profile ↗

65 · GPT-5.6 TerraLow981

Accuracy: 19%
Standard error: ±3.94 pp SE
Evidence: Model profile ↗

65 · GPT-5.1Low981

Accuracy: 19%
Standard error: ±3.94 pp SE
Evidence: Model profile ↗

65 · qwen3.6 Max PreviewNone981

Accuracy: 19%
Standard error: ±3.94 pp SE
Evidence: Model profile ↗

65 · Gemini 3.5 Flash-LiteLow981

Accuracy: 19%
Standard error: ±3.94 pp SE
Evidence: Model profile ↗

70 · qwen3.6 FlashNone977

Accuracy: 18%
Standard error: ±3.86 pp SE
Evidence: Model profile ↗

70 · GLM-5.2Minimal977

Accuracy: 18%
Standard error: ±3.86 pp SE
Evidence: Model profile ↗

70 · GLM-5.2None977

Accuracy: 18%
Standard error: ±3.86 pp SE
Evidence: Model profile ↗

70 · qwen3.5 397b a17bNone977

Accuracy: 18%
Standard error: ±3.86 pp SE
Evidence: Model profile ↗

70 · GPT-5.5None977

Accuracy: 18%
Standard error: ±3.86 pp SE
Evidence: Model profile ↗

70 · Kimi k2.6Not specified977

Accuracy: 18%
Standard error: ±3.86 pp SE
Evidence: Model profile ↗

76 · Claude Sonnet 4.548k972

Accuracy: 17%
Standard error: ±3.78 pp SE
Evidence: Model profile ↗

76 · qwen3.7 PlusNone972

Accuracy: 17%
Standard error: ±3.78 pp SE
Evidence: Model profile ↗

76 · DeepSeek V4 ProNone972

Accuracy: 17%
Standard error: ±3.78 pp SE
Evidence: Model profile ↗

76 · qwen3.7 PlusNot specified972

Accuracy: 17%
Standard error: ±3.78 pp SE
Evidence: Model profile ↗

76 · GPT-5.6 LunaLow972

Accuracy: 17%
Standard error: ±3.78 pp SE
Evidence: Model profile ↗

76 · GPT-5.4Low972

Accuracy: 17%
Standard error: ±3.78 pp SE
Evidence: Model profile ↗

76 · Mistral Large 4 0High972

Accuracy: 17%
Standard error: ±3.78 pp SE
Evidence: Model profile ↗

76 · Qwen3.5-122B-A10BNone972

Accuracy: 17%
Standard error: ±3.78 pp SE
Evidence: Model profile ↗

76 · qwen3.5 PlusNone972

Accuracy: 17%
Standard error: ±3.78 pp SE
Evidence: Model profile ↗

85 · GPT-5Low968

Accuracy: 16%
Standard error: ±3.68 pp SE
Evidence: Model profile ↗

85 · GPT-5.1Medium968

Accuracy: 16%
Standard error: ±3.68 pp SE
Evidence: Model profile ↗

85 · Claude Sonnet 5Not specified968

Accuracy: 16%
Standard error: ±3.68 pp SE
Evidence: Model profile ↗

85 · qwen3.5 FlashNone968

Accuracy: 16%
Standard error: ±3.68 pp SE
Evidence: Model profile ↗

85 · GPT-5.4None968

Accuracy: 16%
Standard error: ±3.68 pp SE
Evidence: Model profile ↗

85 · qwen3.5 PlusNot specified968

Accuracy: 16%
Standard error: ±3.68 pp SE
Evidence: Model profile ↗

85 · Claude Sonnet 4.6Low968

Accuracy: 16%
Standard error: ±3.68 pp SE
Evidence: Model profile ↗

92 · GPT-5.1None963

Accuracy: 15%
Standard error: ±3.59 pp SE
Evidence: Model profile ↗

92 · GPT-5Medium963

Accuracy: 15%
Standard error: ±3.59 pp SE
Evidence: Model profile ↗

92 · qwen3.7 FlashNot specified963

Accuracy: 15%
Standard error: ±3.59 pp SE
Evidence: Model profile ↗

92 · Claude Opus 4.6Not specified963

Accuracy: 15%
Standard error: ±3.59 pp SE
Evidence: Model profile ↗

92 · GLM-5.2Medium963

Accuracy: 15%
Standard error: ±3.59 pp SE
Evidence: Model profile ↗

97 · Muse Spark 1.3xHigh959

Accuracy: 14.14%
Standard error: ±3.52 pp SE
Evidence: Model profile ↗

97 · GPT-5.6 TerraNone959

Accuracy: 14%
Standard error: ±3.49 pp SE
Evidence: Model profile ↗

97 · GPT-5.2None959

Accuracy: 14%
Standard error: ±3.49 pp SE
Evidence: Model profile ↗

97 · GPT-5Minimal959

Accuracy: 14%
Standard error: ±3.49 pp SE
Evidence: Model profile ↗

97 · qwen3.7 FlashNone959

Accuracy: 14%
Standard error: ±3.49 pp SE
Evidence: Model profile ↗

97 · Claude Sonnet 4.6Not specified959

Accuracy: 14%
Standard error: ±3.49 pp SE
Evidence: Model profile ↗

103 · Claude 4.7Not specified954

Accuracy: 13%
Standard error: ±3.38 pp SE
Evidence: Model profile ↗

104 · qwen3.6 FlashNot specified950

Accuracy: 12%
Standard error: ±3.27 pp SE
Evidence: Model profile ↗

104 · Kimi k2.6None950

Accuracy: 12%
Standard error: ±3.27 pp SE
Evidence: Model profile ↗

104 · GPT-5.6 LunaMedium950

Accuracy: 12%
Standard error: ±3.27 pp SE
Evidence: Model profile ↗

104 · Kimi K3Low950

Accuracy: 12%
Standard error: ±3.27 pp SE
Evidence: Model profile ↗

104 · GPT-4 0613Not specified950

Accuracy: 12%
Standard error: ±3.27 pp SE
Evidence: Model profile ↗

104 · Gemini 3.5 Flash-LiteMinimal950

Accuracy: 12%
Standard error: ±3.27 pp SE
Evidence: Model profile ↗

104 · qwen3.6 PlusNone950

Accuracy: 12%
Standard error: ±3.27 pp SE
Evidence: Model profile ↗

104 · GPT-4o-miniNot specified950

Accuracy: 12%
Standard error: ±3.27 pp SE
Evidence: Model profile ↗

112 · GPT-5.4 MiniNone945

Accuracy: 11%
Standard error: ±3.14 pp SE
Evidence: Model profile ↗

113 · GPT-5.2Low941

Accuracy: 10%
Standard error: ±3.02 pp SE
Evidence: Model profile ↗

113 · GPT-5 MiniMinimal941

Accuracy: 10%
Standard error: ±3.02 pp SE
Evidence: Model profile ↗

113 · GPT-5.6 TerraMedium941

Accuracy: 10%
Standard error: ±3.02 pp SE
Evidence: Model profile ↗

116 · Qwen3 235B A22B Thinking 2507Not specified936

Accuracy: 9%
Standard error: ±2.88 pp SE
Evidence: Model profile ↗

116 · GPT-5.4 NanoNone936

Accuracy: 9%
Standard error: ±2.88 pp SE
Evidence: Model profile ↗

116 · GPT-5 NanoLow936

Accuracy: 9%
Standard error: ±2.88 pp SE
Evidence: Model profile ↗

116 · GPT-5 NanoMedium936

Accuracy: 9%
Standard error: ±2.88 pp SE
Evidence: Model profile ↗

120 · GPT-5.4 MiniLow932

Accuracy: 8%
Standard error: ±2.73 pp SE
Evidence: Model profile ↗

120 · GPT-5 NanoHigh932

Accuracy: 8%
Standard error: ±2.73 pp SE
Evidence: Model profile ↗

120 · MiniMax M3None932

Accuracy: 8%
Standard error: ±2.73 pp SE
Evidence: Model profile ↗

120 · GLM 5.3 FlashMax932

Accuracy: 8%
Standard error: ±2.73 pp SE
Evidence: Model profile ↗

124 · GPT-5.4 MiniMedium927

Accuracy: 7%
Standard error: ±2.56 pp SE
Evidence: Model profile ↗

124 · GPT-6 LunaMax927

Accuracy: 7%
Standard error: ±2.56 pp SE
Evidence: Model profile ↗

124 · qwen3.6 27bNone927

Accuracy: 7%
Standard error: ±2.56 pp SE
Evidence: Model profile ↗

124 · MiniMax M3Not specified927

Accuracy: 7%
Standard error: ±2.56 pp SE
Evidence: Model profile ↗

124 · GPT-4.1-MiniNot specified927

Accuracy: 7%
Standard error: ±2.56 pp SE
Evidence: Model profile ↗

124 · Claude Opus 4.6Low927

Accuracy: 7%
Standard error: ±2.56 pp SE
Evidence: Model profile ↗

124 · o3-miniHigh927

Accuracy: 7%
Standard error: ±2.56 pp SE
Evidence: Model profile ↗

131 · GPT-5.4 NanoMedium923

Accuracy: 6%
Standard error: ±2.39 pp SE
Evidence: Model profile ↗

131 · Inkling SmallxHigh923

Accuracy: 6%
Standard error: ±2.39 pp SE
Evidence: Model profile ↗

133 · GPT-5.4 NanoHigh918

Accuracy: 5%
Standard error: ±2.19 pp SE
Evidence: Model profile ↗

133 · o4-miniHigh918

Accuracy: 5%
Standard error: ±2.19 pp SE
Evidence: Model profile ↗

133 · GPT-5 NanoMinimal918

Accuracy: 5%
Standard error: ±2.19 pp SE
Evidence: Model profile ↗

133 · qwen3 Max 2025 09 23Not specified918

Accuracy: 5%
Standard error: ±2.19 pp SE
Evidence: Model profile ↗

133 · GPT-5 MiniHigh918

Accuracy: 5%
Standard error: ±2.19 pp SE
Evidence: Model profile ↗

138 · GPT-5 MiniMedium914

Accuracy: 4%
Standard error: ±1.97 pp SE
Evidence: Model profile ↗

139 · GPT-5.4 NanoLow909

Accuracy: 3%
Standard error: ±1.71 pp SE
Evidence: Model profile ↗

139 · GPT-3.5 Turbo 0125Not specified909

Accuracy: 3%
Standard error: ±1.71 pp SE
Evidence: Model profile ↗

141 · GPT-Oss 120bMedium905

Accuracy: 2%
Standard error: ±1.41 pp SE
Evidence: Model profile ↗

142 · GPT-Oss 120bHigh896

Accuracy: 0%
Standard error: ±0 pp SE
Evidence: Model profile ↗

Source & conversion

Published results by Epoch AI on 100 undisclosed game puzzles.

Mean accuracy and publisher standard error. Resource measurements are unavailable.

How the Lumina conversion works

1,000 + 75 × ((original score − 23.226347986911367) / 16.68350779624929). The result is rounded once. Scores may tie.

Frozen initial 142 official Epoch internal runs; mean accuracy, explicit effort suffixes only. Unknown historical harness versions are not imputed.

Calibration: mystery-game-puzzles-z75-v1 · experimental
Reference mean: 23.226347986911367
Population SD: 16.68350779624929
Coverage: Contributes through the Lumina Reasoning Index.

Identity, exclusions & attribution

Exact model versions and explicit settings only. Unknown settings remain unknown.

The official CSV does not identify per-run harness versions; none are inferred.

Epoch AI methodology · Data and attribution

Publisher terms. Scores and resource figures are attributed to the original publisher.