Benchmark Almanac

CursorBench

Claude Opus 5.5 57.8% leads since Sep 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 2026-03-11 with edit, refactor and bugfix tasks. Versions 3.1 (2026-05-19) and 3.2 (2026-07-08) 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 Mar 2026 Fixing real repositories Tasks kept private Last in a system card Mar 2026 Last model added Oct 2026 Scores refreshed 31 Oct 2026
Level
Expert Specialists with years of training, often over hours of work
Scored
Graded by a model A language model scores each response, usually against a rubric.
Answers
Code A program or patch run against tests.
Leaderboard cursor.com
19
Models scored in Epoch AI data
1
System card reports it, from 1 lab

Claude Opus 5.5 leads with 57.8%

Best score per model in Epoch AI’s data. Where a model ran with several settings or harnesses, we keep its best.

19 models. Scroll sideways for the rest.

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

Labs report scores for 1 model in their cards. We plot every source in the charts below.

6 models have held number one since its release in Mar 2026

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

Cursor Anthropic OpenAI
Epoch AI runs 6 leaders
Claude Opus 5.5 57.8%

Scores over time

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

Run or collected by Epoch AI Reported by the lab in its model card
0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Composer 2 0% 20% 40% 60% 80% 100% 2026 Near the ceiling (85%) Composer 2

Labs report it only in system cards

We read the headline results of 479 launch posts from 51 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
25 Mar 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 xAI Anthropic
CursorBench 4.0 5 leaders
Claude Opus 5.5 57.8%
CursorBench 1 leader
Composer 2 61.3%
CursorBench 3.2 Replaced by 4.0, Sep 2026
Claude Fable 5.1 73.4%
CursorBench 4.0 2 leaders
Claude Opus 5.5 57.8%
Replaces Cursor Bench (2025 internal version) Replaced by CursorBench 3.1

More in Coding

All 243 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%
  • SWE-bench Multilingual In use 300 GitHub-issue tasks across nine programming languages in SWE-bench format. Patch graded by repository tests. Claude Opus 5.5 93.9%
  • 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%
  • 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