Skip to content

Benchmark

DeepSWE

Coding benchmarks compare models on programming tasks. Scores are percentages.

Models measured
36
Scores recorded
36
Sources
2
IQ weight
2.49

Score distribution

55.3%
Mean
59.3%
Median
16.5%
Std dev
30.5%
10th percentile
72.8%
90th percentile
1.6%
Min
1.6%75.5%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.49 for chat models, measured over 36 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model
1Kimi K3.1 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim75.5%
2GPT-6 Astra· DeepSWE74.1%
3Gemini 3.8 Flash· DeepSWE73.8%
4Claude Opus 5· DeepSWE73.7%
5GPT-5.6 Sol· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
72.7%
6Claude Fable 5· DeepSWE69.9%
7GPT-5.6 Terra· DeepSWE
Values from 2 sources
69.6%
8GLM 5.3· DeepSWE69%
9Kimi K3· DeepSWE
Values from 2 sources
68.5%
10Grok 4.6· DeepSWE67.5%
11GPT-5.6 Luna· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
67.2%
12GPT 5.5· DeepSWE67%
13Gemini 3.7 Flash· DeepSWE65.5%
14Hy4 preview· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim64.3%
15GLM 5.3 Flash· DeepSWE63.4%
16DeepSeek V4 Pro· DeepSWE62.8%
17DeepSeek V4 Pro 0813· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim62.7%
18DeepSeek V4 Flash Vision Exp· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim59.3%
19Claude Opus 4.8· DeepSWE59%
20Qwen3.8 Max· DeepSWE57.5%
21Muse Spark 1.2· DeepSWE54.9%
22DeepSeek V4 Flash 0731· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim54.4%
23Claude Sonnet 5· DeepSWE53.9%
24Grok 4.5· DeepSWE
Values from 2 sources
53.8%
25DeepSeek V4 Flash· DeepSWE53.3%
26Muse Spark 1.1· DeepSWE53.3%
27GPT 5.4· DeepSWE51.8%
28Gemini 3.6 Flash· DeepSWE46.7%
29GLM 5.2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim
Values from 2 sources
46.2%
30Gemini 3.5 Flash· DeepSWE36.1%
31Doubao Seed 2.1 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim32.7%
32Kimi K2.7 Code· DeepSWE30.5%
33Claude Sonnet 4.6· DeepSWE29.9%
34Hy3 (Deep Infra)· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim28%
35Doubao Seed 2.1 Turbo· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim23%
36Gemini 3.1 Pro Preview· DeepSWE11.7%

Tokens the model generated per task, as measured by the source. Comparable within this benchmark only.

Data credits: Data by LLM Stats