CursorBench 3.1

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 Opus 4.8, Opus 4.7, GPT-5.5, Sonnet 4.6, Composer 2.5, and Composer 2 scores against average cost per task.70% CursorBench 3.1 score65%60%55%50%45%40%35%30%$12$10$8$6$4$2$0Average cost per taskComposer 2.5Opus 4.7 xhigh(default)GPT-5.5 medium(default)Composer 2Gemini 3.5 FlashOpus 4.8 high(default)Sonnet 4.6 high(default)Kimi 2.6Kimi 2.5
1Opus 4.7 Max64.8%$11.02
2GPT-5.5 Extra High64.3%$4.37
3Opus 4.8 Max63.8%$7.59
4Composer 2.563.2%$0.55
5GPT-5.5 High62.6%$3.59
6Opus 4.8 Extra High62.1%$6.14
7Opus 4.7 Extra High61.6%$7.11
8Opus 4.7 High59.4%$5.01
9GPT-5.5 Medium59.2%$2.22
10Opus 4.8 High58.4%$4.41
11Opus 4.8 Medium56.6%$3.83
12Opus 4.8 Low54.3%$2.93
13Opus 4.7 Medium52.7%$2.93
14Composer 252.2%$0.56
15Gemini 3.5 Flash49.8%$1.94
16Sonnet 4.6 Max49.0%$3.09
17GPT-5.5 Low48.8%$1.19
18Sonnet 4.6 High48.8%$3.06
19Opus 4.7 Low48.3%$1.87
20Kimi 2.647.6%$1.27
21Sonnet 4.6 Medium46.0%$2.64
22Sonnet 4.6 Low41.5%$1.89
23Kimi 2.531.9%$0.87

Changelog

CursorBench 3.1

  • Introduced problems focused on codebase understanding, bugfinding, planning, and code review.
  • Improved grading criteria for some edit 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, and output) to the tokens it used on each CursorBench 3.1 task, then averaging across tasks. Results are subject to variance; small differences in scores may not be statistically meaningful.