Skip to content

Benchmark

SWE-rebench 2026-04

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

Models measured
18
Scores recorded
18
Sources
1
IQ weight
2.07

Score distribution

44.9%
Mean
46%
Median
9.6%
Std dev
37.7%
10th percentile
57.7%
90th percentile
18.6%
Min
18.6%60.5%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 2.07 for chat models, measured over 18 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model
1GPT 5.5tools, xhigh· SWE-rebench60.5%
2Claude Opus 4.8tools, xhigh· SWE-rebench57.7%
3Claude Opus 4.7tools, high· SWE-rebench57.7%
4GLM 5.2tools, high· SWE-rebench51.2%
5Gemini 3.1 Pro Previewtools· SWE-rebench50.2%
6GLM 5.1tools· SWE-rebench50.2%
7Claude Sonnet 4.6tools· SWE-rebench47.4%
8Kimi K2.6tools· SWE-rebench47.4%
9MiniMax M3tools· SWE-rebench46%
10Gemini 3.5 Flashtools· SWE-rebench45.6%
11Claude Opus 4.6tools, high· SWE-rebench43.7%
12MiMo V2.5 Protools· SWE-rebench42.3%
13DeepSeek V4 Protools, high· SWE-rebench41.6%
14DeepSeek V4 Flashtools, high· SWE-rebench40%
15GLM 4.7tools· SWE-rebench37.7%
16Qwen3.6 27Btools· SWE-rebench37.7%
17Qwen3.6 35B A3Btools· SWE-rebench33%
18Gemma 4 31B ITtools· SWE-rebench18.6%

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