Coding benchmark / Cursor

CursorBench

Lumina Verified

Agentic coding on tasks drawn from real engineering work. Published results, checked by Lumina.

Version 4.0Verified 30 October 202681 published resultsOfficial source

Include Claude Max (6)81 of 81 publisher records

CursorBench · Lumina Score

Publisher results on a fixed Lumina scale. Experimental calibration.

LuminaBench

20 of 81 configurations

Lumina Score vs Cost per Task

Published API-equivalent USD per task.

LuminaBench

8 of 19 families

Lumina Score vs Output Tokens

Published output tokens per task; input and cached tokens are not included.

LuminaBench

8 of 19 families

Lumina Score vs Agent Steps

Published agent steps per task. Steps are not tool calls or generation time.

LuminaBench

8 of 19 families

Publisher-reported averages. Numeric confidence intervals were not published. Missing metrics are omitted from their graph; quality scores remain available.

All results

CursorBench 4.0. Original publisher values and provisional Lumina conversion.
Rank
1	Claude Opus 5.5Max · CursorBench agent	1,161	57.8%	$13.43	218,363	185
2	Claude Opus 5.5xHigh · CursorBench agent	1,146	56%	$6.98	101,083	109
2	Claude Opus 5.5High · CursorBench agent	1,146	56%	$3.97	53,078	68
4	Claude Sonnet 5.5Max · CursorBench agent	1,141	55.5%	$7.05	271,920	170
5	Claude Sonnet 5.5xHigh · CursorBench agent	1,121	53.1%	$2.81	100,158	78
6	Gemini 4 ArgonHigh · CursorBench agent	1,119	52.9%	$4.57	109,500	118
7	Gemini 4 ArgonMedium · CursorBench agent	1,118	52.8%	$4.39	103,600	115
8	Claude Opus 5.5Medium · CursorBench agent	1,116	52.5%	$2.91	37,954	54
9	Claude Fable 5.1Max · CursorBench agent	1,110	51.8%	$17.28	117,236	128
10	Claude Fable 5.1xHigh · CursorBench agent	1,108	51.6%	$13.01	87,294	101
11	Gemini 4 ArgonLow · CursorBench agent	1,100	50.6%	$4.09	94,200	110
12	Muse WonderMax · CursorBench agent	1,098	50.4%	$5.25	49,692	114
13	Claude Fable 5.1High · CursorBench agent	1,088	49.2%	$9.08	58,438	77
14	Claude Haiku 5.5Max · CursorBench agent	1,081	48.4%	$1.12	325,934	163
15	Muse WonderxHigh · CursorBench agent	1,079	48.2%	$4.24	39,564	96
16	Claude Sonnet 5.5High · CursorBench agent	1,076	47.8%	$1.2	37,391	41
17	DeepSeek V4.1 Flash 1017Max · CursorBench agent	1,075	47.7%	$0.98	537,047	187
18	Claude Fable 5.1Medium · CursorBench agent	1,067	46.8%	$7.05	45,411	63
19	Claude Opus 5Max · CursorBench agent	1,065	46.6%	$11.95	85,384	106
20	DeepSeek V4.1 Flash 1017Medium · CursorBench agent	1,063	46.3%	$0.8	398,255	189
20	Grok 4.7xHigh · CursorBench agent	1,063	46.3%	$6.01	70,141	88
22	Muse WonderHigh · CursorBench agent	1,061	46.1%	$3.51	33,612	84
22	Claude Opus 5xHigh · CursorBench agent	1,061	46.1%	$11.43	80,094	103
24	DeepSeek V4.1 Flash 1017High · CursorBench agent	1,057	45.6%	$0.77	385,196	186
25	Claude Fable 5.1Low · CursorBench agent	1,053	45.1%	$5.44	34,795	51
25	Muse WonderMedium · CursorBench agent	1,053	45.2%	$2.82	27,124	75
27	DeepSeek V4.1 Flash 1017Low · CursorBench agent	1,051	44.9%	$0.6	290,651	161
28	Claude Opus 5High · CursorBench agent	1,049	44.7%	$9	61,405	86
29	Claude Haiku 5.5xHigh · CursorBench agent	1,046	44.3%	$0.56	143,813	93
30	Grok 4.7High · CursorBench agent	1,042	43.9%	$4.69	56,382	71
31	Claude Opus 5.5Low · CursorBench agent	1,041	43.7%	$1.17	15,811	28
32	Claude Opus 5Medium · CursorBench agent	1,037	43.3%	$6.94	45,272	72
33	GLM 5.3Max · CursorBench agent	1,031	42.6%	$5.05	96,387	166
34	Claude Haiku 5.5High · CursorBench agent	1,029	42.3%	$0.32	77,057	59
35	GPT-5.6 SolMax · CursorBench agent	1,023	41.7%	$8.23	42,944	99
35	Grok 4.7Medium · CursorBench agent	1,023	41.6%	$3.49	36,683	60
35	Muse Spark 1.3Max · CursorBench agent	1,023	41.6%	$2.64	52,005	98
35	Muse WonderLow · CursorBench agent	1,023	41.7%	$1.91	18,880	56
39	Grok 4.6xHigh · CursorBench agent	1,021	41.4%	$6.1	49,814	56
40	GPT-5.6 TerraMax · CursorBench agent	1,020	41.3%	$5.14	60,814	107
41	Claude Opus 5Low · CursorBench agent	1,015	40.7%	$4.87	31,995	57
42	Grok 4.6High · CursorBench agent	1,012	40.4%	$5.2	41,387	48
43	Gemini 3.8 FlashHigh · CursorBench agent	1,006	39.6%	$4.7	162,565	324
44	Claude Sonnet 5.5Medium · CursorBench agent	1,002	39.2%	$0.52	16,036	22
45	GLM 5.3High · CursorBench agent	992	38%	$3.24	60,031	114
46	GPT-5.6 SolxHigh · CursorBench agent	989	37.7%	$4.4	24,729	55
47	Muse Spark 1.3xHigh · CursorBench agent	988	37.5%	$2.1	40,891	83
48	Gemini 3.8 FlashMedium · CursorBench agent	986	37.3%	$4.06	128,364	290
49	GLM 5.3 FlashMax · CursorBench agent	982	36.8%	$0.39	56,410	118
49	Claude Haiku 5.5Medium · CursorBench agent	982	36.9%	$0.17	42,659	39
51	Grok 4.6Medium · CursorBench agent	976	36.1%	$3.48	24,893	40
52	GPT-5.6 LunaMax · CursorBench agent	974	35.9%	$1.03	87,284	208
53	Claude Sonnet 5.5Low · CursorBench agent	973	35.8%	$0.37	11,668	18
54	GPT-5.6 SolHigh · CursorBench agent	972	35.7%	$2.85	16,174	41
55	Muse WonderMinimal · CursorBench agent	959	34.2%	$0.84	9,761	33
55	Claude Sonnet 5Max · CursorBench agent	959	34.1%	$7.17	149,257	140
57	GPT-5.6 TerraxHigh · CursorBench agent	954	33.6%	$1.81	23,436	43
58	Grok 4.6Low · CursorBench agent	953	33.4%	$2.25	16,307	32
58	Muse Spark 1.3High · CursorBench agent	953	33.4%	$1.66	30,654	69
60	GLM 5.3Low · CursorBench agent	952	33.3%	$2.04	31,983	81
61	Grok 4.7Low · CursorBench agent	950	33.1%	$1.58	15,677	40
62	GPT-5.6 LunaxHigh · CursorBench agent	949	33%	$0.44	40,598	98
63	Muse Spark 1.3Medium · CursorBench agent	946	32.6%	$1.49	27,255	64
64	Claude Sonnet 5xHigh · CursorBench agent	941	32%	$4.55	83,373	102
65	GLM 5.3 FlashHigh · CursorBench agent	933	31.1%	$0.25	35,104	84
65	GPT-5.6 SolMedium · CursorBench agent	933	31.1%	$1.77	10,111	32
67	Claude Haiku 5.5Low · CursorBench agent	931	30.9%	$0.08	23,382	25
68	Claude Sonnet 5High · CursorBench agent	930	30.8%	$3.48	61,146	85
69	GPT-5.6 TerraHigh · CursorBench agent	929	30.7%	$1.11	13,162	33
70	GPT-5.6 LunaHigh · CursorBench agent	918	29.4%	$0.25	23,368	64
71	Muse Spark 1.3Low · CursorBench agent	917	29.3%	$0.93	17,483	47
72	Claude Sonnet 5Medium · CursorBench agent	906	28%	$2.31	39,114	65
73	Composer 2.5Not specified · CursorBench agent	904	27.7%	$0.68	17,347	41
74	GPT-5.6 TerraMedium · CursorBench agent	903	27.6%	$0.64	7,307	25
75	GLM 5.3 FlashLow · CursorBench agent	897	26.9%	$0.15	17,831	58
76	GPT-5.6 TerraLow · CursorBench agent	882	25.2%	$0.52	5,914	23
77	GPT-5.6 SolLow · CursorBench agent	877	24.6%	$0.87	4,885	21
78	Muse Spark 1.3Minimal · CursorBench agent	875	24.3%	$0.56	10,620	34
79	Claude Sonnet 5Low · CursorBench agent	873	24.1%	$1.39	23,772	46
80	GPT-5.6 LunaMedium · CursorBench agent	857	22.2%	$0.08	7,642	32
81	GPT-5.6 LunaLow · CursorBench agent	804	16%	$0.03	3,288	18

1 · Claude Opus 5.5Max1161

CursorBench score: 57.8%
API USD/task: $13.43
Output tokens/task: 218,363
Agent steps/task: 185
Evidence: Model profile ↗

2 · Claude Opus 5.5xHigh1146

CursorBench score: 56%
API USD/task: $6.98
Output tokens/task: 101,083
Agent steps/task: 109
Evidence: Model profile ↗

2 · Claude Opus 5.5High1146

CursorBench score: 56%
API USD/task: $3.97
Output tokens/task: 53,078
Agent steps/task: 68
Evidence: Model profile ↗

4 · Claude Sonnet 5.5Max1141

CursorBench score: 55.5%
API USD/task: $7.05
Output tokens/task: 271,920
Agent steps/task: 170
Evidence: Model profile ↗

5 · Claude Sonnet 5.5xHigh1121

CursorBench score: 53.1%
API USD/task: $2.81
Output tokens/task: 100,158
Agent steps/task: 78
Evidence: Model profile ↗

6 · Gemini 4 ArgonHigh1119

CursorBench score: 52.9%
API USD/task: $4.57
Output tokens/task: 109,500
Agent steps/task: 118
Evidence: Model profile ↗

7 · Gemini 4 ArgonMedium1118

CursorBench score: 52.8%
API USD/task: $4.39
Output tokens/task: 103,600
Agent steps/task: 115
Evidence: Model profile ↗

8 · Claude Opus 5.5Medium1116

CursorBench score: 52.5%
API USD/task: $2.91
Output tokens/task: 37,954
Agent steps/task: 54
Evidence: Model profile ↗

9 · Claude Fable 5.1Max1110

CursorBench score: 51.8%
API USD/task: $17.28
Output tokens/task: 117,236
Agent steps/task: 128
Evidence: Model profile ↗

10 · Claude Fable 5.1xHigh1108

CursorBench score: 51.6%
API USD/task: $13.01
Output tokens/task: 87,294
Agent steps/task: 101
Evidence: Model profile ↗

11 · Gemini 4 ArgonLow1100

CursorBench score: 50.6%
API USD/task: $4.09
Output tokens/task: 94,200
Agent steps/task: 110
Evidence: Model profile ↗

12 · Muse WonderMax1098

CursorBench score: 50.4%
API USD/task: $5.25
Output tokens/task: 49,692
Agent steps/task: 114
Evidence: Model profile ↗

13 · Claude Fable 5.1High1088

CursorBench score: 49.2%
API USD/task: $9.08
Output tokens/task: 58,438
Agent steps/task: 77
Evidence: Model profile ↗

14 · Claude Haiku 5.5Max1081

CursorBench score: 48.4%
API USD/task: $1.12
Output tokens/task: 325,934
Agent steps/task: 163
Evidence: Model profile ↗

15 · Muse WonderxHigh1079

CursorBench score: 48.2%
API USD/task: $4.24
Output tokens/task: 39,564
Agent steps/task: 96
Evidence: Model profile ↗

16 · Claude Sonnet 5.5High1076

CursorBench score: 47.8%
API USD/task: $1.2
Output tokens/task: 37,391
Agent steps/task: 41
Evidence: Model profile ↗

17 · DeepSeek V4.1 Flash 1017Max1075

CursorBench score: 47.7%
API USD/task: $0.98
Output tokens/task: 537,047
Agent steps/task: 187
Evidence: Model profile ↗

18 · Claude Fable 5.1Medium1067

CursorBench score: 46.8%
API USD/task: $7.05
Output tokens/task: 45,411
Agent steps/task: 63
Evidence: Model profile ↗

19 · Claude Opus 5Max1065

CursorBench score: 46.6%
API USD/task: $11.95
Output tokens/task: 85,384
Agent steps/task: 106
Evidence: Model profile ↗

20 · DeepSeek V4.1 Flash 1017Medium1063

CursorBench score: 46.3%
API USD/task: $0.8
Output tokens/task: 398,255
Agent steps/task: 189
Evidence: Model profile ↗

20 · Grok 4.7xHigh1063

CursorBench score: 46.3%
API USD/task: $6.01
Output tokens/task: 70,141
Agent steps/task: 88
Evidence: Model profile ↗

22 · Muse WonderHigh1061

CursorBench score: 46.1%
API USD/task: $3.51
Output tokens/task: 33,612
Agent steps/task: 84
Evidence: Model profile ↗

22 · Claude Opus 5xHigh1061

CursorBench score: 46.1%
API USD/task: $11.43
Output tokens/task: 80,094
Agent steps/task: 103
Evidence: Model profile ↗

24 · DeepSeek V4.1 Flash 1017High1057

CursorBench score: 45.6%
API USD/task: $0.77
Output tokens/task: 385,196
Agent steps/task: 186
Evidence: Model profile ↗

25 · Claude Fable 5.1Low1053

CursorBench score: 45.1%
API USD/task: $5.44
Output tokens/task: 34,795
Agent steps/task: 51
Evidence: Model profile ↗

25 · Muse WonderMedium1053

CursorBench score: 45.2%
API USD/task: $2.82
Output tokens/task: 27,124
Agent steps/task: 75
Evidence: Model profile ↗

27 · DeepSeek V4.1 Flash 1017Low1051

CursorBench score: 44.9%
API USD/task: $0.6
Output tokens/task: 290,651
Agent steps/task: 161
Evidence: Model profile ↗

28 · Claude Opus 5High1049

CursorBench score: 44.7%
API USD/task: $9
Output tokens/task: 61,405
Agent steps/task: 86
Evidence: Model profile ↗

29 · Claude Haiku 5.5xHigh1046

CursorBench score: 44.3%
API USD/task: $0.56
Output tokens/task: 143,813
Agent steps/task: 93
Evidence: Model profile ↗

30 · Grok 4.7High1042

CursorBench score: 43.9%
API USD/task: $4.69
Output tokens/task: 56,382
Agent steps/task: 71
Evidence: Model profile ↗

31 · Claude Opus 5.5Low1041

CursorBench score: 43.7%
API USD/task: $1.17
Output tokens/task: 15,811
Agent steps/task: 28
Evidence: Model profile ↗

32 · Claude Opus 5Medium1037

CursorBench score: 43.3%
API USD/task: $6.94
Output tokens/task: 45,272
Agent steps/task: 72
Evidence: Model profile ↗

33 · GLM 5.3Max1031

CursorBench score: 42.6%
API USD/task: $5.05
Output tokens/task: 96,387
Agent steps/task: 166
Evidence: Model profile ↗

34 · Claude Haiku 5.5High1029

CursorBench score: 42.3%
API USD/task: $0.32
Output tokens/task: 77,057
Agent steps/task: 59
Evidence: Model profile ↗

35 · GPT-5.6 SolMax1023

CursorBench score: 41.7%
API USD/task: $8.23
Output tokens/task: 42,944
Agent steps/task: 99
Evidence: Model profile ↗

35 · Grok 4.7Medium1023

CursorBench score: 41.6%
API USD/task: $3.49
Output tokens/task: 36,683
Agent steps/task: 60
Evidence: Model profile ↗

35 · Muse Spark 1.3Max1023

CursorBench score: 41.6%
API USD/task: $2.64
Output tokens/task: 52,005
Agent steps/task: 98
Evidence: Model profile ↗

35 · Muse WonderLow1023

CursorBench score: 41.7%
API USD/task: $1.91
Output tokens/task: 18,880
Agent steps/task: 56
Evidence: Model profile ↗

39 · Grok 4.6xHigh1021

CursorBench score: 41.4%
API USD/task: $6.1
Output tokens/task: 49,814
Agent steps/task: 56
Evidence: Model profile ↗

40 · GPT-5.6 TerraMax1020

CursorBench score: 41.3%
API USD/task: $5.14
Output tokens/task: 60,814
Agent steps/task: 107
Evidence: Model profile ↗

41 · Claude Opus 5Low1015

CursorBench score: 40.7%
API USD/task: $4.87
Output tokens/task: 31,995
Agent steps/task: 57
Evidence: Model profile ↗

42 · Grok 4.6High1012

CursorBench score: 40.4%
API USD/task: $5.2
Output tokens/task: 41,387
Agent steps/task: 48
Evidence: Model profile ↗

43 · Gemini 3.8 FlashHigh1006

CursorBench score: 39.6%
API USD/task: $4.7
Output tokens/task: 162,565
Agent steps/task: 324
Evidence: Model profile ↗

44 · Claude Sonnet 5.5Medium1002

CursorBench score: 39.2%
API USD/task: $0.52
Output tokens/task: 16,036
Agent steps/task: 22
Evidence: Model profile ↗

45 · GLM 5.3High992

CursorBench score: 38%
API USD/task: $3.24
Output tokens/task: 60,031
Agent steps/task: 114
Evidence: Model profile ↗

46 · GPT-5.6 SolxHigh989

CursorBench score: 37.7%
API USD/task: $4.4
Output tokens/task: 24,729
Agent steps/task: 55
Evidence: Model profile ↗

47 · Muse Spark 1.3xHigh988

CursorBench score: 37.5%
API USD/task: $2.1
Output tokens/task: 40,891
Agent steps/task: 83
Evidence: Model profile ↗

48 · Gemini 3.8 FlashMedium986

CursorBench score: 37.3%
API USD/task: $4.06
Output tokens/task: 128,364
Agent steps/task: 290
Evidence: Model profile ↗

49 · GLM 5.3 FlashMax982

CursorBench score: 36.8%
API USD/task: $0.39
Output tokens/task: 56,410
Agent steps/task: 118
Evidence: Model profile ↗

49 · Claude Haiku 5.5Medium982

CursorBench score: 36.9%
API USD/task: $0.17
Output tokens/task: 42,659
Agent steps/task: 39
Evidence: Model profile ↗

51 · Grok 4.6Medium976

CursorBench score: 36.1%
API USD/task: $3.48
Output tokens/task: 24,893
Agent steps/task: 40
Evidence: Model profile ↗

52 · GPT-5.6 LunaMax974

CursorBench score: 35.9%
API USD/task: $1.03
Output tokens/task: 87,284
Agent steps/task: 208
Evidence: Model profile ↗

53 · Claude Sonnet 5.5Low973

CursorBench score: 35.8%
API USD/task: $0.37
Output tokens/task: 11,668
Agent steps/task: 18
Evidence: Model profile ↗

54 · GPT-5.6 SolHigh972

CursorBench score: 35.7%
API USD/task: $2.85
Output tokens/task: 16,174
Agent steps/task: 41
Evidence: Model profile ↗

55 · Muse WonderMinimal959

CursorBench score: 34.2%
API USD/task: $0.84
Output tokens/task: 9,761
Agent steps/task: 33
Evidence: Model profile ↗

55 · Claude Sonnet 5Max959

CursorBench score: 34.1%
API USD/task: $7.17
Output tokens/task: 149,257
Agent steps/task: 140
Evidence: Model profile ↗

57 · GPT-5.6 TerraxHigh954

CursorBench score: 33.6%
API USD/task: $1.81
Output tokens/task: 23,436
Agent steps/task: 43
Evidence: Model profile ↗

58 · Grok 4.6Low953

CursorBench score: 33.4%
API USD/task: $2.25
Output tokens/task: 16,307
Agent steps/task: 32
Evidence: Model profile ↗

58 · Muse Spark 1.3High953

CursorBench score: 33.4%
API USD/task: $1.66
Output tokens/task: 30,654
Agent steps/task: 69
Evidence: Model profile ↗

60 · GLM 5.3Low952

CursorBench score: 33.3%
API USD/task: $2.04
Output tokens/task: 31,983
Agent steps/task: 81
Evidence: Model profile ↗

61 · Grok 4.7Low950

CursorBench score: 33.1%
API USD/task: $1.58
Output tokens/task: 15,677
Agent steps/task: 40
Evidence: Model profile ↗

62 · GPT-5.6 LunaxHigh949

CursorBench score: 33%
API USD/task: $0.44
Output tokens/task: 40,598
Agent steps/task: 98
Evidence: Model profile ↗

63 · Muse Spark 1.3Medium946

CursorBench score: 32.6%
API USD/task: $1.49
Output tokens/task: 27,255
Agent steps/task: 64
Evidence: Model profile ↗

64 · Claude Sonnet 5xHigh941

CursorBench score: 32%
API USD/task: $4.55
Output tokens/task: 83,373
Agent steps/task: 102
Evidence: Model profile ↗

65 · GLM 5.3 FlashHigh933

CursorBench score: 31.1%
API USD/task: $0.25
Output tokens/task: 35,104
Agent steps/task: 84
Evidence: Model profile ↗

65 · GPT-5.6 SolMedium933

CursorBench score: 31.1%
API USD/task: $1.77
Output tokens/task: 10,111
Agent steps/task: 32
Evidence: Model profile ↗

67 · Claude Haiku 5.5Low931

CursorBench score: 30.9%
API USD/task: $0.08
Output tokens/task: 23,382
Agent steps/task: 25
Evidence: Model profile ↗

68 · Claude Sonnet 5High930

CursorBench score: 30.8%
API USD/task: $3.48
Output tokens/task: 61,146
Agent steps/task: 85
Evidence: Model profile ↗

69 · GPT-5.6 TerraHigh929

CursorBench score: 30.7%
API USD/task: $1.11
Output tokens/task: 13,162
Agent steps/task: 33
Evidence: Model profile ↗

70 · GPT-5.6 LunaHigh918

CursorBench score: 29.4%
API USD/task: $0.25
Output tokens/task: 23,368
Agent steps/task: 64
Evidence: Model profile ↗

71 · Muse Spark 1.3Low917

CursorBench score: 29.3%
API USD/task: $0.93
Output tokens/task: 17,483
Agent steps/task: 47
Evidence: Model profile ↗

72 · Claude Sonnet 5Medium906

CursorBench score: 28%
API USD/task: $2.31
Output tokens/task: 39,114
Agent steps/task: 65
Evidence: Model profile ↗

73 · Composer 2.5Not specified904

CursorBench score: 27.7%
API USD/task: $0.68
Output tokens/task: 17,347
Agent steps/task: 41
Evidence: Model profile ↗

74 · GPT-5.6 TerraMedium903

CursorBench score: 27.6%
API USD/task: $0.64
Output tokens/task: 7,307
Agent steps/task: 25
Evidence: Model profile ↗

75 · GLM 5.3 FlashLow897

CursorBench score: 26.9%
API USD/task: $0.15
Output tokens/task: 17,831
Agent steps/task: 58
Evidence: Model profile ↗

76 · GPT-5.6 TerraLow882

CursorBench score: 25.2%
API USD/task: $0.52
Output tokens/task: 5,914
Agent steps/task: 23
Evidence: Model profile ↗

77 · GPT-5.6 SolLow877

CursorBench score: 24.6%
API USD/task: $0.87
Output tokens/task: 4,885
Agent steps/task: 21
Evidence: Model profile ↗

78 · Muse Spark 1.3Minimal875

CursorBench score: 24.3%
API USD/task: $0.56
Output tokens/task: 10,620
Agent steps/task: 34
Evidence: Model profile ↗

79 · Claude Sonnet 5Low873

CursorBench score: 24.1%
API USD/task: $1.39
Output tokens/task: 23,772
Agent steps/task: 46
Evidence: Model profile ↗

80 · GPT-5.6 LunaMedium857

CursorBench score: 22.2%
API USD/task: $0.08
Output tokens/task: 7,642
Agent steps/task: 32
Evidence: Model profile ↗

81 · GPT-5.6 LunaLow804

CursorBench score: 16%
API USD/task: $0.03
Output tokens/task: 3,288
Agent steps/task: 18
Evidence: Model profile ↗

Source & conversion

Cursor reports solution-correctness scores on coding tasks drawn from real engineering work. The exact version4.0 task count, agent build and dated API model snapshots are not published. Older methodology descriptions should not be treated as the complete version4.0 protocol.

API-equivalent task cost uses published input, cache-read, cache-write and output pricing. Tokens means output tokens per task. Agent steps are publisher-reported averages; a formal step definition and comparable timing data are unavailable.

How the Lumina conversion works

1,000 + 75 × ((original score − 38.952) / 8.773237486811809). The result is rounded once. Scores may tie.

The fixed reference contains75 initial published configurations, excluding six Anthropic Max settings. Population standard deviation uses N. New entries use the same frozen parameters. This is a display calibration, not equivalence to our original creation tests.

Calibration: cursorbench-4-z75-v1 · experimental
Reference mean: 38.952
Population SD: 8.773237486811809
Coverage: Independent benchmark; no contribution to the current Capabilities Index.

Identity, exclusions & attribution

GPT-5.6 Sol, Terra and Luna are distinct from Lumina’s GPT-6.x test models. Serving-provider labels are retained separately from model producers. A source label does not establish a dated API snapshot or a match to a Lumina harness configuration.

All published reasoning settings, including Claude Max, are included by default. Display labels use Low and xHigh; source labels remain in the retained records.

Published leaderboard · Publisher methodology · Model identity and pricing documentation

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