Skip to content

Benchmark

CursorBench 4.0

Coding benchmarks compare models on programming tasks. Scores are percentages. Higher is better.

Models and systems measured
14
Scores recorded
69
Sources
2
IQ weight
0.81

Score distribution

41.1%
Mean
41.6%
Median
10.4%
Std dev
27.7%
10th percentile
56%
90th percentile
16%
Min
16%60%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.81 for chat models, measured over 9 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model or systemScore
51Claude Sonnet 5Cursor production agent · extra high effort · $4.55/task · 83,373 tokens · 102 steps· Reported by Cursor· via BenchBoard docs32%
52GPT-5.6 SolCursor production agent · medium effort · $1.77/task · 10,111 tokens · 32 steps· Reported by Cursor· via BenchBoard docs31.1%
53Claude Sonnet 5Cursor production agent · high effort · $3.48/task · 61,146 tokens · 85 steps· Reported by Cursor· via BenchBoard docs30.8%
54GPT-5.6 TerraCursor production agent · high effort · $1.11/task · 13,162 tokens · 33 steps· Reported by Cursor· via BenchBoard docs30.7%
55GPT-5.6 LunaCursor production agent · high effort · $0.25/task · 23,368 tokens · 64 steps· Reported by Cursor· via BenchBoard docs29.4%
56Muse Spark 1.3Cursor production agent · low effort · $0.93/task · 17,483 tokens · 47 steps· Reported by Cursor· via BenchBoard docs29.3%
57Claude Sonnet 5Cursor production agent · medium effort · $2.31/task · 39,114 tokens · 65 steps· Reported by Cursor· via BenchBoard docs28%
58GPT-5.6 TerraCursor production agent · medium effort · $0.64/task · 7,307 tokens · 25 steps· Reported by Cursor· via BenchBoard docs27.6%
59GPT-5.6 TerraCursor production agent · low effort · $0.52/task · 5,914 tokens · 23 steps· Reported by Cursor· via BenchBoard docs25.2%
60GPT-5.6 SolCursor production agent · low effort · $0.87/task · 4,885 tokens · 21 steps· Reported by Cursor· via BenchBoard docs24.6%
61Muse Spark 1.3Cursor production agent · minimal effort · $0.56/task · 10,620 tokens · 34 steps· Reported by Cursor· via BenchBoard docs24.3%
62Claude Sonnet 5Cursor production agent · low effort · $1.39/task · 23,772 tokens · 46 steps· Reported by Cursor· via BenchBoard docs24.1%
63GPT-5.6 LunaCursor production agent · medium effort · $0.08/task · 7,642 tokens · 32 steps· Reported by Cursor· via BenchBoard docs22.2%
64GPT-5.6 LunaCursor production agent · low effort · $0.03/task · 3,288 tokens · 18 steps· Reported by Cursor· via BenchBoard docs16%

Showing 51–64 of 64 results.

Not ranked

The source marks these results as descriptive, not capability scores, so we list them by name instead of ranking them.

Model or systemScore
Composer 2.5Cursor production agent · default effort · $0.68/task · 17,347 tokens · 41 steps· Reported by Cursor· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.27.7%
Grok 4.6Cursor production agent · extra high effort · $6.1/task · 49,814 tokens · 56 steps· Reported by Cursor· via BenchBoard docs41.4%
Grok 4.6Cursor production agent · high effort · $5.2/task · 41,387 tokens · 48 steps· Reported by Cursor· via BenchBoard docs40.4%
Grok 4.6Cursor production agent · low effort · $2.25/task · 16,307 tokens · 32 steps· Reported by Cursor· via BenchBoard docs33.4%
Grok 4.6Cursor production agent · medium effort · $3.48/task · 24,893 tokens · 40 steps· Reported by Cursor· via BenchBoard docs36.1%

Data credits: BenchBoard by SYuan03 (github.com/SYuan03/benchboard). MIT covers its code and data organization; no results licence is inferred. Numeric relays retain the original stating-source links. · Data by LLM Stats