Benchmark Almanac

CursorBench

Gemini 4 Carbon 60.4% leads since October 2026
Composer 2 61.3% leads since March 2026
Only in system cards

Agents do ambiguous multi-file coding tasks from real Cursor sessions in Cursor's harness. Version 3.0 covers edit, refactor and bugfix problems. Agentic graders score correctness, with cost reported.

Cursor's changelog dates 3.0 to March 11th, 2026 with edit, refactor and bugfix tasks. Versions 3.1 (May 19th, 2026) and 3.2 (July 8th, 2026) changed the task mix, so they have their own records. Cursor's Composer 2 post and report (2026-03) use 3.0.

Made by
Cursor
Released
March 2026
Kind
Fixing real repositories
Level
Expert
Scored
Graded by a model
Answers
Code
Tasks
Kept private
Strongest model tested
Claude Fable 5.2 #1 on the capability index
Last in a system card
March 2026
Last model added
November 2026
Scores refreshed
November 12th, 2026
21
Models scored on a board Epoch AI copied
1
System card reports it, from 1 lab

Gemini 4 Carbon leads with 60.4%

Best score per model from cursor.com, as Epoch AI copied it. We don’t read that board ourselves yet. Where a model ran with several settings, we keep its best.

Composer 2 leads with 61.3%, claims included

Best score per model from every source, including labs’ own numbers and claims Epoch AI collected from reports and papers. Labs pick their own settings, so a claim can beat an independent run of the same model.

21 models. Scroll sideways for the rest.

  1. 60.4 Gemini 4 Carbon
  2. 60.0 Claude Fable 5.2
  3. 57.8 Claude Opus 5.5
  4. 55.5 Claude Sonnet 5.5
  5. 52.9 Gemini 4 Argon
  6. 51.8 Claude Fable 5.1
  7. 50.4 Muse Wonder
  8. 48.4 Claude Haiku 5.5
  9. 47.7 DeepSeek V4.1 Flash 1017
  10. 46.6 Claude Opus 5
  11. 46.3 Grok 4.7
  12. 42.6 GLM-5.3
  13. 41.7 GPT-5.6 Sol
  14. 41.6 Muse Spark 1.3
  15. 41.4 Grok 4.6
  16. 41.3 GPT-5.6 Terra
  17. 39.6 Gemini 3.8 Flash
  18. 36.8 GLM-5.3-Flash
  19. 35.9 GPT-5.6 Luna
  20. 34.1 Claude Sonnet 5
  21. 27.7 Composer 2.5
  1. 1 Gemini 4 Carbon high 60.4%
  2. 2 Claude Fable 5.2 max 60%
  3. 3 Claude Opus 5.5 max 57.8%
  4. 4 Claude Sonnet 5.5 max 55.5%
  5. 5 Gemini 4 Argon high 52.9%
  6. 6 Claude Fable 5.1 max 51.8%
  7. 7 Muse Wonder max 50.4%
  8. 8 Claude Haiku 5.5 max 48.4%
  9. 9 DeepSeek V4.1 Flash 1017 max 47.7%
  10. 10 Claude Opus 5 max 46.6%
  11. 11 Grok 4.7 xhigh 46.3%
  12. 12 GLM-5.3 max 42.6%
  13. 13 GPT-5.6 Sol max 41.7%
  14. 14 Muse Spark 1.3 max 41.6%
  15. 15 Grok 4.6 xhigh 41.4%
  16. 16 GPT-5.6 Terra max 41.3%
  17. 17 Gemini 3.8 Flash high 39.6%
  18. 18 GLM-5.3-Flash max 36.8%
  19. 19 GPT-5.6 Luna max 35.9%
  20. 20 Claude Sonnet 5 max 34.1%
  21. 21 Composer 2.5 27.7%

Source: Cursor.com, via Epoch AI (CC BY 4.0)

22 models. Scroll sideways for the rest.

  1. 61.3 Composer 2
  2. 60.4 Gemini 4 Carbon
  3. 60.0 Claude Fable 5.2
  4. 57.8 Claude Opus 5.5
  5. 55.5 Claude Sonnet 5.5
  6. 52.9 Gemini 4 Argon
  7. 51.8 Claude Fable 5.1
  8. 50.4 Muse Wonder
  9. 48.4 Claude Haiku 5.5
  10. 47.7 DeepSeek V4.1 Flash 1017
  11. 46.6 Claude Opus 5
  12. 46.3 Grok 4.7
  13. 42.6 GLM-5.3
  14. 41.7 GPT-5.6 Sol
  15. 41.6 Muse Spark 1.3
  16. 41.4 Grok 4.6
  17. 41.3 GPT-5.6 Terra
  18. 39.6 Gemini 3.8 Flash
  19. 36.8 GLM-5.3-Flash
  20. 35.9 GPT-5.6 Luna
  21. 34.1 Claude Sonnet 5
  22. 27.7 Composer 2.5
  1. 1 Composer 2 61.3%
  2. 2 Gemini 4 Carbon high 60.4%
  3. 3 Claude Fable 5.2 max 60%
  4. 4 Claude Opus 5.5 max 57.8%
  5. 5 Claude Sonnet 5.5 max 55.5%
  6. 6 Gemini 4 Argon high 52.9%
  7. 7 Claude Fable 5.1 max 51.8%
  8. 8 Muse Wonder max 50.4%
  9. 9 Claude Haiku 5.5 max 48.4%
  10. 10 DeepSeek V4.1 Flash 1017 max 47.7%
  11. 11 Claude Opus 5 max 46.6%
  12. 12 Grok 4.7 xhigh 46.3%
  13. 13 GLM-5.3 max 42.6%
  14. 14 GPT-5.6 Sol max 41.7%
  15. 15 Muse Spark 1.3 max 41.6%
  16. 16 Grok 4.6 xhigh 41.4%
  17. 17 GPT-5.6 Terra max 41.3%
  18. 18 Gemini 3.8 Flash high 39.6%
  19. 19 GLM-5.3-Flash max 36.8%
  20. 20 GPT-5.6 Luna max 35.9%
  21. 21 Claude Sonnet 5 max 34.1%
  22. 22 Composer 2.5 27.7%

Sources: Labs’ model cards and cursor.com, via Epoch AI (CC BY 4.0)

#ModelScoreSettingDateSource
1 Gemini 4 Carbon 60.4% high Oct 2026 cursor.com, via Epoch AI
2 Claude Fable 5.2 60% max Nov 2026 cursor.com, via Epoch AI
3 Claude Opus 5.5 57.8% max Sep 2026 cursor.com, via Epoch AI
4 Claude Sonnet 5.5 55.5% max Sep 2026 cursor.com, via Epoch AI
5 Gemini 4 Argon 52.9% high Sep 2026 cursor.com, via Epoch AI
6 Claude Fable 5.1 51.8% max Sep 2026 cursor.com, via Epoch AI
7 Muse Wonder 50.4% max Oct 2026 cursor.com, via Epoch AI
8 Claude Haiku 5.5 48.4% max Oct 2026 cursor.com, via Epoch AI
9 DeepSeek V4.1 Flash 1017 47.7% max Oct 2026 cursor.com, via Epoch AI
10 Claude Opus 5 46.6% max Jul 2026 cursor.com, via Epoch AI
11 Grok 4.7 46.3% xhigh Sep 2026 cursor.com, via Epoch AI
12 GLM-5.3 42.6% max Aug 2026 cursor.com, via Epoch AI
13 GPT-5.6 Sol 41.7% max Jul 2026 cursor.com, via Epoch AI
14 Muse Spark 1.3 41.6% max Sep 2026 cursor.com, via Epoch AI
15 Grok 4.6 41.4% xhigh Aug 2026 cursor.com, via Epoch AI
16 GPT-5.6 Terra 41.3% max Jul 2026 cursor.com, via Epoch AI
17 Gemini 3.8 Flash 39.6% high Sep 2026 cursor.com, via Epoch AI
18 GLM-5.3-Flash 36.8% max Aug 2026 cursor.com, via Epoch AI
19 GPT-5.6 Luna 35.9% max Jul 2026 cursor.com, via Epoch AI
20 Claude Sonnet 5 34.1% max Jun 2026 cursor.com, via Epoch AI
21 Composer 2.5 27.7% May 2026 cursor.com, via Epoch AI
#ModelScoreSettingDateSource
1 Composer 2 61.3% Mar 2026 Lab model card
2 Gemini 4 Carbon 60.4% high Oct 2026 cursor.com, via Epoch AI
3 Claude Fable 5.2 60% max Nov 2026 cursor.com, via Epoch AI
4 Claude Opus 5.5 57.8% max Sep 2026 cursor.com, via Epoch AI
5 Claude Sonnet 5.5 55.5% max Sep 2026 cursor.com, via Epoch AI
6 Gemini 4 Argon 52.9% high Sep 2026 cursor.com, via Epoch AI
7 Claude Fable 5.1 51.8% max Sep 2026 cursor.com, via Epoch AI
8 Muse Wonder 50.4% max Oct 2026 cursor.com, via Epoch AI
9 Claude Haiku 5.5 48.4% max Oct 2026 cursor.com, via Epoch AI
10 DeepSeek V4.1 Flash 1017 47.7% max Oct 2026 cursor.com, via Epoch AI
11 Claude Opus 5 46.6% max Jul 2026 cursor.com, via Epoch AI
12 Grok 4.7 46.3% xhigh Sep 2026 cursor.com, via Epoch AI
13 GLM-5.3 42.6% max Aug 2026 cursor.com, via Epoch AI
14 GPT-5.6 Sol 41.7% max Jul 2026 cursor.com, via Epoch AI
15 Muse Spark 1.3 41.6% max Sep 2026 cursor.com, via Epoch AI
16 Grok 4.6 41.4% xhigh Aug 2026 cursor.com, via Epoch AI
17 GPT-5.6 Terra 41.3% max Jul 2026 cursor.com, via Epoch AI
18 Gemini 3.8 Flash 39.6% high Sep 2026 cursor.com, via Epoch AI
19 GLM-5.3-Flash 36.8% max Aug 2026 cursor.com, via Epoch AI
20 GPT-5.6 Luna 35.9% max Jul 2026 cursor.com, via Epoch AI
21 Claude Sonnet 5 34.1% max Jun 2026 cursor.com, via Epoch AI
22 Composer 2.5 27.7% May 2026 cursor.com, via Epoch AI

Labs report scores for 1 model in their cards. Switch the scores above to rank them together.

7 models have held number one on cursor.com since its release in March 2026 Number one over time

Lab claims count from the day the card came out, independent runs from the model’s release.

Cursor Anthropic OpenAI Google DeepMind
cursor.com, via Epoch AI 7 leaders
Gemini 4 Carbon 60.4%

Source: Cursor.com, via Epoch AI (CC BY 4.0)

Scores over time

All the scores we have, one dot each, so a model tested at several settings appears more than once.

Reported by the lab in its model card On another leaderboard, collected by Epoch AI
0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Composer 2 0% 20% 40% 60% 80% 100% 2026 2027 Near the ceiling (85%) Composer 2

Sources: Other boards, via Epoch AI (CC BY 4.0) and labs’ model cards

Labs report it only in system cards

We read the headline results of 522 launch posts from 52 labs, and none of them list this benchmark. Some system cards do, below.

One system card reports it

Besides launch posts, we read the system cards, model cards and technical reports labs publish with a release. They report far more, including safety and frontier-risk tests.

DateDocumentScoreAs the document writes itSection
Mar 25, 2026 Composer 2 technical report 61.3 CursorBench (CursorBench-3), Cursor harness (Table 1 and 7.1) Capabilities

Versions

We switch the top row to each new version when it comes out. Open it for one row per version. An old version stays in play while people still test it, so whoever first takes it past 85% gets credit for closing it out.

Cursor Anthropic OpenAI Google DeepMind xAI
CursorBench 4.0 12 leaders
Gemini 4 Carbon 60.4%
CursorBench 7 leaders
Gemini 4 Carbon 60.4%
CursorBench 3.2 Replaced by 4.0, Sep 2026
Claude Fable 5.1 73.4%
CursorBench 4.0 3 leaders
Gemini 4 Carbon 60.4%

Sources: Official leaderboards, boards copied by Epoch AI (CC BY 4.0) and labs’ model cards

Replaces Cursor Bench (2025 internal version) Replaced by CursorBench 3.1

More in Coding

All 282 in Coding
  • SWE-bench Verified In use 500 human-validated GitHub issues from 12 Python repos. Agent writes a patch that must pass hidden tests. Claude Opus 4.7 83.5%
  • LiveCodeBench v6 In use Competitive programming problems collected after model cutoffs from LeetCode, AtCoder, Codeforces. Scored pass@1 on hidden tests. DeepSeek-V4-Pro 93.5%
  • SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Fable 5.2 96.6%
  • DeepSWE In use 113 original long-horizon engineering tasks across 91 repos and five languages, written from scratch. Hand-written verifiers accept alternative solutions. GPT-6 Astra 74.1%
  • SciCode In use Scientist-curated research coding problems split into subproblems across natural sciences. Code checked against test cases. Seed2.1 Pro 59.8%
  • Codeforces rating In use Model solves recent Codeforces contest problems. An Elo-style rating is computed from simulated contest performance. DeepSeek V4.1 Flash 3,471