Skip to content

Benchmark

SWE-rebench 2026-05

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

Models measured
7
Scores recorded
7
Sources
1
IQ weight
0.95

Score distribution

40.5%
Mean
41.6%
Median
13.2%
Std dev
28.2%
10th percentile
50.7%
90th percentile
19.1%
Min
19.1%61.1%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.95 for chat models, measured over 7 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#Model
1GLM 5.2tools, high· SWE-rebench61.1%
2MiMo V2.5 Protools· SWE-rebench50.7%
3MiniMax M3tools· SWE-rebench48%
4DeepSeek V4 Protools, high· SWE-rebench41.6%
5Qwen3.6 27Btools· SWE-rebench34.7%
6Qwen3.6 35B A3Btools· SWE-rebench28.2%
7Qwen3.5 35B A3Btools· SWE-rebench19.1%

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