Cursor · CursorBench

We evaluate agents on ambiguous, multi-file tasks from real Cursor sessions. Higher scores are better.More about CursorBenchA scatter and line chart comparing Fable 5, Opus 4.8, Opus 4.7, GPT-5.5, Sonnet 5, Sonnet 4.6, GLM 5.2, Composer 2.5, and Composer 2 scores against average cost per task.75% CursorBench 3.1 score70%65%60%55%50%45%$20$16$12$8$4$0Average cost per taskFable 5 highComposer 2.5GPT-5.5 mediumGemini 3.5 FlashOpus 4.8 highSonnet 5 highKimi K2.7 CodeGLM 5.2 highModel1Fable 5 Max72.9%$18.0263,842762Fable 5 Extra High72.0%$13.7448,754633Fable 5 High70.6%$10.8137,173544Fable 5 Medium69.8%$8.2728,507475Opus 4.7 Max64.8%$11.0262,989966GPT-5.5 Extra High64.3%$4.3717,905467Fable 5 Low64.2%$5.7018,882368Opus 4.8 Max63.8%$7.5977,370609Composer 2.563.2%$0.5515,1523710GPT-5.5 High62.6%$3.5913,3294011Opus 4.8 Extra

High62.1%$6.1455,6225412Opus 4.7 Extra High61.6%$7.1143,9427213Sonnet 5 Max61.2%$6.8793,4859314Opus 4.7 High59.4%$5.0132,2275915GPT-5.5 Medium59.2%$2.229,0653516Opus 4.8 High58.4%$4.4136,7884517Sonnet 5 Extra High58.4%$5.2358,2288618Sonnet 5 High57.0%$3.7441,7356619Opus 4.8 Medium56.6%$3.8331,6844120Sonnet 5 Medium54.9%$2.5727,4695321GLM 5.2 Max54.6%$3.1151,3128322Opus 4.8 Low54.3%$2.9322,7263623Opus 4.7 Medium52.7%$2.9319,1934124Kimi K2.7 Code52.7%$1.9232,9027025Composer 252.2%$0.5614,1634026GLM 5.2 High50.7%$2.4630,6217627Gemini

3.5 Flash49.8%$1.9435,1057928Sonnet 4.6 Max49.0%$3.0940,2805529GPT-5.5 Low48.8%$1.194,9232430Sonnet 4.6 High48.8%$3.0637,3525731Opus 4.7 Low48.3%$1.8713,1642932Sonnet 5 Low47.7%$1.4617,0283733Kimi 2.647.6%$1.2724,7835634Sonnet 4.6 Medium46.0%$2.6431,3605035Sonnet 4.6 Low41.5%$1.8921,2115036Kimi 2.531.9%$0.879,44630ChangelogCursorBench 3.1Introduced problems focused on codebase understanding, bugfinding, planning, and code review.Improved grading criteria for some edit tasks.CursorBench 3.0Initial 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 CursorBench 3.1 task, then averaging across tasks. Results are subject to variance; small differences in scores may not be statistically meaningful.