activeCodingAgents

CursorBench

Metric: Score (higher is better)Introduced: 2026

The 42 rows below are LLMReference's July 18, 2026 snapshot of CursorBench 3.2. CursorBench is a vendor-run evaluation of IDE-native, multi-file coding-agent workflows, not standalone base models, and its public harness is not independently reproducible. Each reasoning or effort configuration remains a separate result; do not collapse these rows into a neutral per-model winner.

Configurations

42

tracked in CursorBench 3.2

Score band

70.5 – 37.6

best → lowest tracked

Snapshot trend

—

CursorBench 3.2 · need ≥2 snapshots

Published snapshot

CursorBench 3.2 · observed Jul 18, 2026 · 42 configurations.

CursorBench 3.2 configurations are published exactly as Cursor reports them, including cost, token, and step averages. No per-model maximum is selected. Grok 4.5 rows carry Cursor's training-data-advantage disclosure and are not neutral winner evidence.

Compare candidates

#Published configuration and provenanceRelative scoreScore

1

Configuration: Fable 5 Max

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $17.32Tokens/task: 103,525Steps/task: 72Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

70.5

2

Configuration: Fable 5 Extra High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $11.73Tokens/task: 64,971Steps/task: 56Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

68.4

3

Configuration: GPT-5.6 Sol Max

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $5.69Tokens/task: 28,320Steps/task: 48Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

67.2

4

Configuration: Grok 4.5 High*

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $1.51Tokens/task: 19,521Steps/task: 33Source

Qualification: Cursor discloses a training-data advantage; do not use this result for a neutral ranking claim.

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

66.7

5

Configuration: Fable 5 High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $8.77Tokens/task: 43,747Steps/task: 48Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

66.5

6

Configuration: Grok 4.5 Medium*

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $1.54Tokens/task: 18,914Steps/task: 34Source

Qualification: Cursor discloses a training-data advantage; do not use this result for a neutral ranking claim.

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

65.4

7

Configuration: Fable 5 Medium

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $6.80Tokens/task: 30,366Steps/task: 41Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

65.2

8

Configuration: GPT-5.6 Terra Max

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $2.89Tokens/task: 32,969Steps/task: 47Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

64.9

9

Configuration: GPT-5.6 Sol Extra High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $3.88Tokens/task: 19,699Steps/task: 38Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

64.5

10

Configuration: Grok 4.5 Low*

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $1.22Tokens/task: 15,841Steps/task: 31Source

Qualification: Cursor discloses a training-data advantage; do not use this result for a neutral ranking claim.

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

63.5

11

Configuration: GPT-5.6 Sol High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $2.79Tokens/task: 13,867Steps/task: 32Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

63.5

12

Claude Opus 4.8

Configuration: Opus 4.8 Max

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $5.77Tokens/task: 71,411Steps/task: 44Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

62.3

13

Configuration: Fable 5 Low

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $4.46Tokens/task: 18,182Steps/task: 31Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

62.1

14

Claude Sonnet 5

Configuration: Sonnet 5 Max

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $6.45Tokens/task: 92,882Steps/task: 86Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

61.5

15

Configuration: GPT-5.6 Luna Max

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $1.97Tokens/task: 87,973Steps/task: 61Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

61.1

16

Configuration: GPT-5.6 Sol Medium

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $1.95Tokens/task: 9,747Steps/task: 27Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

60.0

17

Claude Opus 4.8

Configuration: Opus 4.8 Extra High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $4.50Tokens/task: 51,121Steps/task: 40Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

59.4

18

Configuration: GPT-5.6 Terra Extra High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $1.44Tokens/task: 16,089Steps/task: 29Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

59.2

19

Claude Sonnet 5

Configuration: Sonnet 5 Extra High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $4.16Tokens/task: 52,871Steps/task: 67Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

58.7

20

Configuration: GPT-5.5 High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $2.05Tokens/task: 12,183Steps/task: 28Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

58.4

21

Configuration: GPT-5.5 Extra High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $2.85Tokens/task: 17,534Steps/task: 32Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

58.4

22

Claude Opus 4.8

Configuration: Opus 4.8 High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $3.15Tokens/task: 33,548Steps/task: 33Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

58.0

23

Configuration: GPT-5.6 Luna Extra High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $1.14Tokens/task: 22,480Steps/task: 48Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

57.7

24

Claude Sonnet 5

Configuration: Sonnet 5 High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $3.19Tokens/task: 39,483Steps/task: 57Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

56.9

25

Configuration: GPT-5.6 Luna High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $0.82Tokens/task: 15,141Steps/task: 40Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

56.8

26

Claude Opus 4.8

Configuration: Opus 4.8 Medium

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $2.81Tokens/task: 28,384Steps/task: 32Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

56.1

27

Configuration: Composer 2.5

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $0.44Tokens/task: 14,286Steps/task: 33Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

56.1

28

Configuration: GLM 5.2 Max

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $1.76Tokens/task: 35,946Steps/task: 58Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

55.0

29

Configuration: GPT-5.6 Terra High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $0.89Tokens/task: 9,468Steps/task: 23Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

54.2

30

Configuration: GPT-5.5 Medium

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $1.51Tokens/task: 8,522Steps/task: 25Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

53.8

31

Claude Opus 4.8

Configuration: Opus 4.8 Low

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $2.02Tokens/task: 19,624Steps/task: 27Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

53.1

32

Configuration: GPT-5.6 Sol Low

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $1.01Tokens/task: 5,104Steps/task: 19Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

52.6

33

Claude Sonnet 5

Configuration: Sonnet 5 Medium

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $2.16Tokens/task: 26,200Steps/task: 46Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

52.4

34

Configuration: GLM 5.2 High

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $1.19Tokens/task: 21,829Steps/task: 49Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

51.5

35

Configuration: GPT-5.6 Terra Medium

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $0.61Tokens/task: 6,222Steps/task: 20Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

50.3

36

Configuration: Kimi K2.7 Code

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $1.43Tokens/task: 31,247Steps/task: 58Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

49.7

37

Gemini 3.5 Flash

Configuration: Gemini 3.5 Flash

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $2.20Tokens/task: 46,702Steps/task: 77Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

48.8

38

Configuration: GPT-5.6 Luna Medium

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $0.39Tokens/task: 7,095Steps/task: 28Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

47.7

39

Claude Sonnet 5

Configuration: Sonnet 5 Low

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $1.30Tokens/task: 16,269Steps/task: 33Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

47.7

40

Configuration: GPT-5.6 Terra Low

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $0.53Tokens/task: 5,312Steps/task: 19Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

46.9

41

Configuration: GPT-5.5 Low

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $0.98Tokens/task: 5,168Steps/task: 20Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

46.6

42

Configuration: GPT-5.6 Luna Low

Version: CursorBench 3.2Harness: CursorBench 3.2 productized Cursor-agent workflowEvaluator: CursorObserved: Jul 18, 2026Confidence: confirmedCost/task: $0.16Tokens/task: 3,209Steps/task: 17Source

Notes: Cursor vendor-reported result; not independently reproducible. Results are subject to variance, and small score differences may not be statistically meaningful.

37.6

Other version snapshots

Every benchmark version stays in a separate dated table. Scores, ranks, and changes are not compared across versions.

CursorBench 3.2, Grok 4.6 High (xAI first-party evals table)

Observed Aug 12, 2026 · 1 configuration.

#Published configuration and provenanceRelative scoreScore

1

Configuration: Grok 4.6 High

Version: CursorBench 3.2, Grok 4.6 High (xAI first-party evals table)Harness: Not recordedEvaluator: Not recordedObserved: Aug 12, 2026Confidence: Not recordedSource

Notes: xAI launch post evals table reports Grok 4.6 High at 69.9% on CursorBench v3.2.

69.9

CursorBench 3.1

Observed Jun 30, 2026 · 12 configurations.

#Published configuration and provenanceRelative scoreScore

1

Configuration: Fable 5 Max

Version: CursorBench 3.1Harness: CursorBench 3.1Evaluator: CursorObserved: Jun 30, 2026Confidence: confirmedSource

Notes: Highest CursorBench 3.1 score across Cursor's published effort configurations for this base model.

72.9

2

Claude Opus 4.7

Configuration: Opus 4.7 Max

Version: CursorBench 3.1Harness: CursorBench 3.1Evaluator: CursorObserved: Jun 30, 2026Confidence: confirmedSource

Notes: Highest CursorBench 3.1 score across Cursor's published effort configurations for this base model.

64.8

3

Configuration: GPT-5.5 Extra High

Version: CursorBench 3.1Harness: CursorBench 3.1Evaluator: CursorObserved: Jun 30, 2026Confidence: confirmedSource

Notes: Highest CursorBench 3.1 score across Cursor's published effort configurations for this base model.

64.3

4

Claude Opus 4.8

Configuration: Opus 4.8 Max

Version: CursorBench 3.1Harness: CursorBench 3.1Evaluator: CursorObserved: Jun 30, 2026Confidence: confirmedSource

Notes: Highest CursorBench 3.1 score across Cursor's published effort configurations for this base model.

63.8

5

Configuration: Composer 2.5 (single reported configuration)

Version: CursorBench 3.1Harness: CursorBench 3.1Evaluator: CursorObserved: Jun 30, 2026Confidence: confirmedSource

Notes: Cursor published one CursorBench 3.1 configuration for this model; no cross-effort selection was needed.

63.2

6

Claude Sonnet 5

Configuration: Sonnet 5 Max

Version: CursorBench 3.1Harness: CursorBench 3.1Evaluator: CursorObserved: Jun 30, 2026Confidence: confirmedSource

Notes: Highest CursorBench 3.1 score across Cursor's published effort configurations for this base model.

61.2

7

Configuration: GLM 5.2 Max

Version: CursorBench 3.1Harness: CursorBench 3.1Evaluator: CursorObserved: Jun 30, 2026Confidence: confirmedSource

Notes: Highest CursorBench 3.1 score across Cursor's published effort configurations for this base model.

54.6

8

Configuration: Composer 2 (single reported configuration)

Version: CursorBench 3.1Harness: CursorBench 3.1Evaluator: CursorObserved: Jun 30, 2026Confidence: confirmedSource

Notes: Cursor published one CursorBench 3.1 configuration for this model; no cross-effort selection was needed.

52.2

9

Gemini 3.5 Flash

Configuration: Gemini 3.5 Flash (single reported configuration)

Version: CursorBench 3.1Harness: CursorBench 3.1Evaluator: CursorObserved: Jun 30, 2026Confidence: confirmedSource

Notes: Cursor published one CursorBench 3.1 configuration for this model; no cross-effort selection was needed.

49.8

10

Claude Sonnet 4.6

Configuration: Sonnet 4.6 Max

Version: CursorBench 3.1Harness: CursorBench 3.1Evaluator: CursorObserved: Jun 30, 2026Confidence: confirmedSource

Notes: Highest CursorBench 3.1 score across Cursor's published effort configurations for this base model.

49.0

11

Configuration: Kimi 2.6 (single reported configuration)

Version: CursorBench 3.1Harness: CursorBench 3.1Evaluator: CursorObserved: Jun 30, 2026Confidence: confirmedSource

Notes: Cursor published one CursorBench 3.1 configuration for this model; no cross-effort selection was needed.

47.6

12

Configuration: Kimi 2.5 (single reported configuration)

Version: CursorBench 3.1Harness: CursorBench 3.1Evaluator: CursorObserved: Jun 30, 2026Confidence: confirmedSource

Notes: Cursor published one CursorBench 3.1 configuration for this model; no cross-effort selection was needed.

31.9

How to read this benchmark

This benchmark scores models where higher is better. Use scores for directional filtering and shortlisting, not universal quality ranking; then validate pricing, context window, provider availability, and fit for your workload.

Trust this score when

The row names the exact model configuration, benchmark version, evaluator, harness, and observation date.
Cost, token, and step averages are evaluated together with the score for the same configuration.
You validate the configuration in your own Cursor workload before making a model choice.

Be cautious when

Rows are collapsed into one score per base model or compared across benchmark versions.
Small score differences are treated as decisive despite Cursor's variance caveat.
A qualified row, including Grok 4.5's disclosed training-data advantage, is used as neutral winner evidence.

Related benchmarks

Mostly Basic Programming Problems

Mostly Basic Programming Problems+

Last reviewed: Jul 18, 2026

Resources