Skip to content

Benchmark

CursorBench

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

Models measured
14
Scores recorded
55
Sources
2
IQ weight
0.00

Score distribution

2.8%
Mean
0.46%
Median
12.2%
Std dev
0.287%
10th percentile
0.614%
90th percentile
0.213%
Min
0.213%70%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.00 for chat models, measured over 14 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model
1Claude Opus 4.7Cursor internal evaluation· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.70%
2Gemini 4 NeonExtra High / xhigh· Epoch AI Benchmarking Hub data65.4%
3Gemini 4 NeonHigh· Epoch AI Benchmarking Hub data63.7%
4Gemini 4 CarbonHigh· Epoch AI Benchmarking Hub data61.5%
5Gemini 4 HydrogenExtra High / xhigh· Epoch AI Benchmarking Hub data61.4%
6Claude Fable 5.2Max· Epoch AI Benchmarking Hub data61.3%
7Claude Sonnet 5Cursor production Agent harness· Reported by Anthropic· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.61.2%
8Gemini 4 CarbonMedium· Epoch AI Benchmarking Hub data60.4%
9Claude Fable 5.2Extra High / xhigh· Epoch AI Benchmarking Hub data58.8%
10Claude Opus 5.5Max· Epoch AI Benchmarking Hub data58.1%
11Gemini 4 NeonMedium· Epoch AI Benchmarking Hub data57.7%
12Gemini 4 HydrogenHigh· Epoch AI Benchmarking Hub data57.3%
13Claude Opus 5.5Extra High / xhigh· Epoch AI Benchmarking Hub data56.7%
14Claude Fable 5.2High· Epoch AI Benchmarking Hub data56.5%
15Gemini 4 ArgonHigh· Epoch AI Benchmarking Hub data56.5%
16Gemini 4 ArgonMedium· Epoch AI Benchmarking Hub data55.8%
17Gemini 4 CarbonLow· Epoch AI Benchmarking Hub data55%
18Claude Opus 5.5High· Epoch AI Benchmarking Hub data54.5%
19Claude Sonnet 5.5Max· Epoch AI Benchmarking Hub data54.4%
20Claude Fable 5.2Medium· Epoch AI Benchmarking Hub data54.2%
21Claude Fable 5.2Low· Epoch AI Benchmarking Hub data53.4%
22Claude Opus 5.5Medium· Epoch AI Benchmarking Hub data50%
23Claude Sonnet 5.5Extra High / xhigh· Epoch AI Benchmarking Hub data49.9%
24Grok 4.7Extra High / xhigh· Epoch AI Benchmarking Hub data48.6%
25Claude Haiku 5.5Max· Epoch AI Benchmarking Hub data48.4%
26Gemini 4 ArgonLow· Epoch AI Benchmarking Hub data47.9%
27Gemini 4 HydrogenMedium· Epoch AI Benchmarking Hub data46.4%
28Grok 4.7High· Epoch AI Benchmarking Hub data46%
29Gemini 4 NeonLow· Epoch AI Benchmarking Hub data45.5%
30Claude Haiku 5.5Extra High / xhigh· Epoch AI Benchmarking Hub data45.2%
31Grok 4.6Extra High / xhigh· Epoch AI Benchmarking Hub data44.3%
32Claude Sonnet 5.5High· Epoch AI Benchmarking Hub data43.8%
33Claude Opus 5.5Low· Epoch AI Benchmarking Hub data42.7%
34Grok 4.7Medium· Epoch AI Benchmarking Hub data42.5%
35Muse Spark 1.3Max· Epoch AI Benchmarking Hub data41.9%
36Grok 4.6High· Epoch AI Benchmarking Hub data41.3%
37GPT-5.6 SolMax· Epoch AI Benchmarking Hub data40.9%
38Grok 4.6Medium· Epoch AI Benchmarking Hub data38.9%
39Claude Haiku 5.5High· Epoch AI Benchmarking Hub data37.5%
40GPT-5.6 SolExtra High / xhigh· Epoch AI Benchmarking Hub data37.4%
41Muse Spark 1.3Extra High / xhigh· Epoch AI Benchmarking Hub data34.6%
42Claude Sonnet 5.5Medium· Epoch AI Benchmarking Hub data34.3%
43Claude Haiku 5.5Medium· Epoch AI Benchmarking Hub data34.2%
44GPT-5.6 SolHigh· Epoch AI Benchmarking Hub data33.2%
45Grok 4.6Low· Epoch AI Benchmarking Hub data33.1%
46Muse Spark 1.3High· Epoch AI Benchmarking Hub data32.1%
47Claude Sonnet 5.5Low· Epoch AI Benchmarking Hub data31.8%
48Grok 4.7Low· Epoch AI Benchmarking Hub data31.5%
49Muse Spark 1.3Medium· Epoch AI Benchmarking Hub data31.3%
50GPT-5.6 SolMedium· Epoch AI Benchmarking Hub data28.7%

Showing 1–50 of 55 results.

Tokens per task, input and output together, as measured by the source. Comparable within this benchmark only.

Data credits: Epoch AI Benchmarking Hub, CC BY 4.0 (licence) · 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.