CursorBench 4.1

We evaluate agents on ambiguous, multi-file tasks from real Cursor sessions. Higher scores are better.

More about CursorBench
A scatter and line chart comparing Fable 5.2, Opus 5.5, Grok 4.7, Grok 4.6, GPT-5.6 Sol, Sonnet 5.5, Haiku 5.5, Gemini 4 Neon, Gemini 4 Hydrogen, Gemini 4 Carbon, Gemini 4 Argon, and Muse Spark 1.3 scores against average cost per task.70%CursorBench 4.1 score65%60%55%50%45%40%35%30%25%20%$18$15$12$9$6$3$0Average cost per taskGemini 4 NeonGemini 4 CarbonGemini 4HydrogenFable 5.2Opus 5.5Sonnet 5.5Haiku 5.5Muse Spark 1.3GPT-5.6 SolGrok 4.7
Model
1Gemini 4 Neon Extra High65.4%$10.67136,27373
2Gemini 4 Neon High63.7%$6.5488,59656
3Gemini 4 Carbon High61.5%$8.52114,51897
4Gemini 4 Hydrogen Extra High61.4%$8.09205,981113
5Fable 5.2 Max61.3%$48.11541,003295
6Gemini 4 Carbon Medium60.4%$7.22103,18983
7Fable 5.2 Extra High58.8%$17.84194,918122
8Opus 5.5 Max58.1%$11.88213,070144
9Gemini 4 Neon Medium57.7%$3.0747,05941
10Gemini 4 Hydrogen High57.3%$4.57113,80777
11Opus 5.5 Extra High56.7%$6.54104,90195
12Fable 5.2 High56.5%$10.51108,21377
13Gemini 4 Argon High56.5%$9.14135,038118
14Gemini 4 Argon Medium55.8%$7.64128,231101
15Gemini 4 Carbon Low55.0%$8.6448,526127
16Opus 5.5 High54.5%$3.8756,16264
17Sonnet 5.5 Max54.4%$8.63342,273199
18Fable 5.2 Medium54.2%$5.0147,67840
19Fable 5.2 Low53.4%$4.2640,38434
20Opus 5.5 Medium50.0%$2.8539,73652
21Sonnet 5.5 Extra High49.9%$3.13112,29987
22Grok 4.7 Extra High48.6%$10.1990,268110
23Haiku 5.5 Max48.4%$1.14344,890166
24Gemini 4 Argon Low47.9%$7.9640,541121
25Gemini 4 Hydrogen Medium46.4%$1.2843,60645
26Grok 4.7 High46.0%$6.8865,85084
27Gemini 4 Neon Low45.5%$1.2619,58026
28Haiku 5.5 Extra High45.2%$0.56149,31494
29Grok 4.6 Extra High44.3%$5.2658,55462
30Sonnet 5.5 High43.8%$1.4143,08848
31Opus 5.5 Low42.7%$1.4319,49732
32Grok 4.7 Medium42.5%$5.5241,17370
33Muse Spark 1.3 Max41.9%$2.8956,572113
34Grok 4.6 High41.3%$4.2846,33052
35GPT-5.6 Sol Max40.9%$7.6040,466100
36Grok 4.6 Medium38.9%$2.5926,37440
37Haiku 5.5 High37.5%$0.3177,19661
38GPT-5.6 Sol Extra High37.4%$4.3223,92158
39Muse Spark 1.3 Extra High34.6%$2.3347,14296
40Sonnet 5.5 Medium34.3%$0.6319,14426
41Haiku 5.5 Medium34.2%$0.1743,44141
42GPT-5.6 Sol High33.2%$3.0016,31944
43Grok 4.6 Low33.1%$1.7616,86333
44Muse Spark 1.3 High32.1%$1.9736,56285
45Sonnet 5.5 Low31.8%$0.4413,79821
46Grok 4.7 Low31.5%$2.7314,49342
47Muse Spark 1.3 Medium31.3%$1.7532,03477
48GPT-5.6 Sol Medium28.7%$1.9710,55434
49Haiku 5.5 Low27.8%$0.0823,67327
50Muse Spark 1.3 Low27.5%$1.1320,67358
51Gemini 4 Hydrogen Low23.9%$0.2810,49722
52GPT-5.6 Sol Low23.0%$0.985,20622
53Muse Spark 1.3 Minimal21.3%$0.6912,78942

Changelog

Tasks

  • CursorBench 4.1
    • Modified instruction following problems.
    • Increased agent time limit.

Reporting

  • Updated Sonnet 5.5 results to account for adjusted pricing.

Tasks

  • CursorBench 4.0
    • Introduced new long-horizon problems focused on edit, refactor, investigation, intent understanding, managing jobs, and design adherence.

Reporting

  • Updated Sonnet 5 results to account for adjusted pricing.

Reporting

  • Updated GPT-5.6 Terra and Luna results to account for adjusted pricing.

Reporting

  • Updated GPT-5.6 Sol, Terra, and Luna results to account for cache write costs.

Tasks

  • CursorBench 3.2
    • Introduced instruction following and advanced tool use problems.

Tasks

  • CursorBench 3.1
    • Introduced problems focused on codebase understanding, bugfinding, planning, and code review.
    • Improved grading criteria for some edit tasks.

Tasks

  • CursorBench 3.0
    • Initial set of tasks focused on edit, refactor, and bugfix problems.

Avg cost / task is computed by applying each model's published per-million-token pricing (input, cache read, cache write, and output) to the tokens it used on each task. Results are subject to variance; small differences in scores may not be statistically meaningful.