Skip to content

Benchmark

SWE-Marathon

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

Models measured
13
Scores recorded
18
Sources
2
IQ weight
0.77

Score distribution

35.8%
Mean
42%
Median
13.7%
Std dev
16%
10th percentile
53.1%
90th percentile
13%
Min
13%57.5%
The band covers the middle 80% of scores. The line marks the median.

This benchmark has a weight of 0.77 for chat models, measured over 13 models. The weight is how much it counts toward llminfo IQ. We measure it from the data, not from the raw scores.

Leaderboard

#ModelScore
1Muse Wonder Full· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim57.5%
2Muse Wonder· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim53.1%
3Muse Wonder 1.0· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim53.1%
4Kimi K3.1 Pro· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim51.7%
5Grok 4.7· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim46%
6GLM 5.3· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim42.5%
7GLM 5.3v1.1 · Claude Code 2.1.207 · max · temp 1.0 · top_p .95 · 1M · 128K output· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.42.5%
8Kimi K3· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim42%
9Kimi K3H20-calibrated pre-v1.1 task snapshot · max effort · temp 1.0 · top_p 1.0· Reported by Moonshot AI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.42%
10Hy4 preview· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim31.9%
11Hy4 previewofficial model-card appendix· Reported by Tencent Hy· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.31.9%
12Grok 4.5· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim29%
13Grok 4.5resolution rate · pass@1· Reported by SpaceXAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.29%
14Claude Opus 4.8max· Reported by SpaceXAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.26%
15Claude Fable 5max· Reported by SpaceXAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.24%
16Claude Opus 4.7max· Reported by SpaceXAI· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.16%
17GLM 5.2· LLM Stats data· Reported by the model's developer. The source relaying it did not verify it. claim13%
18GLM 5.21M context · max effort · 128K output· Reported by Z.ai· via BenchBoard docs· Reported by the model's developer. The source relaying it did not verify it.13%

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