Multi-SWE-bench

1,632 repository issues spanning 7 languages (Java, TypeScript, JavaScript, Go, Rust, C, C++).26 Oct 2026

Sources

Multi-SWE-bench: A Multilingual Benchmark for Issue Resolving — published result table ↗
Daoguang Zan, Zhirong Huang, Wei Liu, Hanwu Chen, Linhao Zhang, Shulin Xin et al. · 2025
All 19 authors
Daoguang Zan, Zhirong Huang, Wei Liu, Hanwu Chen, Linhao Zhang, Shulin Xin, Lu Chen, Qi Liu, Xiaojian Zhong, Aoyan Li, Siyao Liu, Yongsheng Xiao, Liangqiang Chen, Yuyu Zhang, Jing Su, Tianyu Liu, Rui Long, Kai Shen, Liang Xiang
Dates & evaluation details
Multi-SWE-bench · Resolved · Python · language-specific results · 2026-10-26
Published model comparison, 2025. Multi-SWE-bench · language-specific results. Resolved · Python; Multi-SWE-bench · language-specific results. Model configurations and missing observations follow the original publication. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Multi-SWE-bench · Resolved · Java · language-specific results · 2026-10-26
Published model comparison, 2025. Multi-SWE-bench · language-specific results. Resolved · Java; Multi-SWE-bench · language-specific results. Model configurations and missing observations follow the original publication. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Multi-SWE-bench · Resolved · TypeScript · language-specific results · 2026-10-26
Published model comparison, 2025. Multi-SWE-bench · language-specific results. Resolved · TypeScript; Multi-SWE-bench · language-specific results. Model configurations and missing observations follow the original publication. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Multi-SWE-bench · Resolved · JavaScript · language-specific results · 2026-10-26
Published model comparison, 2025. Multi-SWE-bench · language-specific results. Resolved · JavaScript; Multi-SWE-bench · language-specific results. Model configurations and missing observations follow the original publication. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Multi-SWE-bench · Resolved · Go · language-specific results · 2026-10-26
Published model comparison, 2025. Multi-SWE-bench · language-specific results. Resolved · Go; Multi-SWE-bench · language-specific results. Model configurations and missing observations follow the original publication. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Multi-SWE-bench · Resolved · Rust · language-specific results · 2026-10-26
Published model comparison, 2025. Multi-SWE-bench · language-specific results. Resolved · Rust; Multi-SWE-bench · language-specific results. Model configurations and missing observations follow the original publication. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Multi-SWE-bench · Resolved · C · language-specific results · 2026-10-26
Published model comparison, 2025. Multi-SWE-bench · language-specific results. Resolved · C; Multi-SWE-bench · language-specific results. Model configurations and missing observations follow the original publication. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.
Multi-SWE-bench · Resolved · C++ · language-specific results · 2026-10-26
Published model comparison, 2025. Multi-SWE-bench · language-specific results. Resolved · C++; Multi-SWE-bench · language-specific results. Model configurations and missing observations follow the original publication. Table 4; source precision and any reported uncertainty retained. These are source-reported observations, not evaluations run by RLTasks.

TRACE THE SCORE · Per-model notes use published breakdowns where available.

What it measures

1,632 repository issues spanning 7 languages (Java, TypeScript, JavaScript, Go, Rust, C, C++).

Reading the measurement

Look beyond a plausible code sample to the executable artifact. Tests, repository context, language coverage and time limits shape the result. Separate first-attempt correctness from success after debugging, retries or agent assistance.

Published evaluation

Resolved · Python; Multi-SWE-bench · language-specific results. Model configurations and missing observations follow the original publication.

Where this fits in an evaluation suite

The agent wrapper is part of the system. Pin tools, dependencies, timeouts, tests, and the repository revision. Compare contest coding separately from repository maintenance.

Start with a capability hypothesis, then choose a second test with a different failure mode. Keep the checkpoint, reasoning effort, agent tools, sampling policy, and scoring procedure fixed when measuring a training intervention. Record failures and missing outputs in the denominator.

A reproducible comparison

Resolved · Python; Multi-SWE-bench · language-specific results. Model configurations and missing observations follow the original publication.

Publish the exact run configuration, item coverage, score units, and available uncertainty. If a source supplies only an aggregate or a single run, preserve that limit. A new model name or a rounded score alone does not establish a capability improvement.

PRIMARY EVIDENCE & FURTHER READING

Multi-SWE-bench: A Multilingual Benchmark for Issue Resolving — published result table ↗
Daoguang Zan, Zhirong Huang, Wei Liu, Hanwu Chen, Linhao Zhang, Shulin Xin et al. · 2025
All 19 authors
Daoguang Zan, Zhirong Huang, Wei Liu, Hanwu Chen, Linhao Zhang, Shulin Xin, Lu Chen, Qi Liu, Xiaojian Zhong, Aoyan Li, Siyao Liu, Yongsheng Xiao, Liangqiang Chen, Yuyu Zhang, Jing Su, Tianyu Liu, Rui Long, Kai Shen, Liang Xiang

Reviewed 2026-10-26. Explanation and interpretation by RLTasks. Benchmark ownership remains with its original authors.

Continue exploring

All code & engineering benchmarks →

		Evidence	What the result shows
Claude-3.7-Sonnet →AnthropicMulti-SWE-bench · language-specific results · MSWE-agent	8.59% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MSWE-agent	Leads this displayed comparison (tied). Source notes Source cell: 8.59
Claude-3.7-Sonnet →AnthropicMulti-SWE-bench · language-specific results · MopenHands	8.59% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MopenHands	Leads this displayed comparison (tied). Source notes Source cell: 8.59
Claude-3.5-Sonnet →AnthropicMulti-SWE-bench · language-specific results · MSWE-agent	4.69% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MSWE-agent	#3 of 27 in this displayed comparison; 3.9 pp below the leader. Source notes Source cell: 4.69
OpenAI-o1 →OpenAIMulti-SWE-bench · language-specific results · MSWE-agent	3.91% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MSWE-agent	#4 of 27 in this displayed comparison; 4.68 pp below the leader. Source notes Source cell: 3.91
OpenAI-o3-mini-high →OpenAIMulti-SWE-bench · language-specific results · MagentLess	3.91% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MagentLess	#4 of 27 in this displayed comparison; 4.68 pp below the leader. Source notes Source cell: 3.91
Claude-3.5-Sonnet →AnthropicMulti-SWE-bench · language-specific results · MopenHands	3.13% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MopenHands	#6 of 27 in this displayed comparison; 5.46 pp below the leader. Source notes Source cell: 3.13
DeepSeek V3 →DeepSeekMulti-SWE-bench · language-specific results · MagentLess	3.13% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MagentLess	#6 of 27 in this displayed comparison; 5.46 pp below the leader. Source notes Source cell: 3.13
DeepSeek V3 →DeepSeekMulti-SWE-bench · language-specific results · MopenHands	3.13% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MopenHands	#6 of 27 in this displayed comparison; 5.46 pp below the leader. Source notes Source cell: 3.13
OpenAI-o1 →OpenAIMulti-SWE-bench · language-specific results · MopenHands	3.13% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MopenHands	#6 of 27 in this displayed comparison; 5.46 pp below the leader. Source notes Source cell: 3.13
Claude-3.7-Sonnet →AnthropicMulti-SWE-bench · language-specific results · MagentLess	2.34% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MagentLess	#10 of 27 in this displayed comparison; 6.25 pp below the leader. Source notes Source cell: 2.34
DeepSeek V3 →DeepSeekMulti-SWE-bench · language-specific results · MSWE-agent	2.34% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MSWE-agent	#10 of 27 in this displayed comparison; 6.25 pp below the leader. Source notes Source cell: 2.34
DeepSeek-R1 →DeepSeekMulti-SWE-bench · language-specific results · MopenHands	2.34% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MopenHands	#10 of 27 in this displayed comparison; 6.25 pp below the leader. Source notes Source cell: 2.34
Doubao-1.5-pro →ByteDanceMulti-SWE-bench · language-specific results · MSWE-agent	2.34% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MSWE-agent	#10 of 27 in this displayed comparison; 6.25 pp below the leader. Source notes Source cell: 2.34
OpenAI-o3-mini-high →OpenAIMulti-SWE-bench · language-specific results · MSWE-agent	2.34% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MSWE-agent	#10 of 27 in this displayed comparison; 6.25 pp below the leader. Source notes Source cell: 2.34
Claude-3.5-Sonnet →AnthropicMulti-SWE-bench · language-specific results · MagentLess	1.56% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MagentLess	#15 of 27 in this displayed comparison; 7.03 pp below the leader. Source notes Source cell: 1.56
GPT-4o →OpenAIMulti-SWE-bench · language-specific results · MagentLess	1.56% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MagentLess	#15 of 27 in this displayed comparison; 7.03 pp below the leader. Source notes Source cell: 1.56
GPT-4o →OpenAIMulti-SWE-bench · language-specific results · MSWE-agent	1.56% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MSWE-agent	#15 of 27 in this displayed comparison; 7.03 pp below the leader. Source notes Source cell: 1.56
OpenAI-o1 →OpenAIMulti-SWE-bench · language-specific results · MagentLess	1.56% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MagentLess	#15 of 27 in this displayed comparison; 7.03 pp below the leader. Source notes Source cell: 1.56
OpenAI-o3-mini-high →OpenAIMulti-SWE-bench · language-specific results · MopenHands	1.56% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MopenHands	#15 of 27 in this displayed comparison; 7.03 pp below the leader. Source notes Source cell: 1.56
Qwen2.5-72B-Instruct →QwenMulti-SWE-bench · language-specific results · MSWE-agent	1.56% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MSWE-agent	#15 of 27 in this displayed comparison; 7.03 pp below the leader. Source notes Source cell: 1.56
DeepSeek-R1 →DeepSeekMulti-SWE-bench · language-specific results · MagentLess	0.78% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MagentLess	#21 of 27 in this displayed comparison; 7.81 pp below the leader. Source notes Source cell: 0.78
DeepSeek-R1 →DeepSeekMulti-SWE-bench · language-specific results · MSWE-agent	0.78% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MSWE-agent	#21 of 27 in this displayed comparison; 7.81 pp below the leader. Source notes Source cell: 0.78
Qwen2.5-72B-Instruct →QwenMulti-SWE-bench · language-specific results · MagentLess	0.78% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MagentLess	#21 of 27 in this displayed comparison; 7.81 pp below the leader. Source notes Source cell: 0.78
Qwen2.5-72B-Instruct →QwenMulti-SWE-bench · language-specific results · MopenHands	0.78% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MopenHands	#21 of 27 in this displayed comparison; 7.81 pp below the leader. Source notes Source cell: 0.78
Doubao-1.5-pro →ByteDanceMulti-SWE-bench · language-specific results · MagentLess	0% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MagentLess	#25 of 27 in this displayed comparison; 8.59 pp below the leader. Source notes Source cell: 0.00
Doubao-1.5-pro →ByteDanceMulti-SWE-bench · language-specific results · MopenHands	0% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MopenHands	#25 of 27 in this displayed comparison; 8.59 pp below the leader. Source notes Source cell: 0.00
GPT-4o →OpenAIMulti-SWE-bench · language-specific results · MopenHands	0% 1 report	Multi-SWE-bench · Resolved · C[1]Multi-SWE-bench · language-specific results · MopenHands	#25 of 27 in this displayed comparison; 8.59 pp below the leader. Source notes Source cell: 0.00
Claude-3.7-Sonnet →AnthropicMulti-SWE-bench · language-specific results · MopenHands	14.73% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MopenHands	Leads this displayed comparison. Source notes Source cell: 14.73
Claude-3.5-Sonnet →AnthropicMulti-SWE-bench · language-specific results · MopenHands	12.4% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MopenHands	#2 of 27 in this displayed comparison; 2.33 pp below the leader. Source notes Source cell: 12.40
Claude-3.7-Sonnet →AnthropicMulti-SWE-bench · language-specific results · MSWE-agent	11.63% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MSWE-agent	#3 of 27 in this displayed comparison; 3.1 pp below the leader. Source notes Source cell: 11.63
DeepSeek V3 →DeepSeekMulti-SWE-bench · language-specific results · MSWE-agent	7.75% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MSWE-agent	#4 of 27 in this displayed comparison; 6.98 pp below the leader. Source notes Source cell: 7.75
DeepSeek V3 →DeepSeekMulti-SWE-bench · language-specific results · MopenHands	7.75% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MopenHands	#4 of 27 in this displayed comparison; 6.98 pp below the leader. Source notes Source cell: 7.75
Claude-3.5-Sonnet →AnthropicMulti-SWE-bench · language-specific results · MSWE-agent	6.98% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MSWE-agent	#6 of 27 in this displayed comparison; 7.75 pp below the leader. Source notes Source cell: 6.98
GPT-4o →OpenAIMulti-SWE-bench · language-specific results · MagentLess	6.98% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MagentLess	#6 of 27 in this displayed comparison; 7.75 pp below the leader. Source notes Source cell: 6.98
OpenAI-o3-mini-high →OpenAIMulti-SWE-bench · language-specific results · MopenHands	6.98% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MopenHands	#6 of 27 in this displayed comparison; 7.75 pp below the leader. Source notes Source cell: 6.98
DeepSeek-R1 →DeepSeekMulti-SWE-bench · language-specific results · MSWE-agent	6.2% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MSWE-agent	#9 of 27 in this displayed comparison; 8.53 pp below the leader. Source notes Source cell: 6.20
Doubao-1.5-pro →ByteDanceMulti-SWE-bench · language-specific results · MSWE-agent	6.2% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MSWE-agent	#9 of 27 in this displayed comparison; 8.53 pp below the leader. Source notes Source cell: 6.20
OpenAI-o1 →OpenAIMulti-SWE-bench · language-specific results · MagentLess	5.43% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MagentLess	#11 of 27 in this displayed comparison; 9.3 pp below the leader. Source notes Source cell: 5.43
OpenAI-o3-mini-high →OpenAIMulti-SWE-bench · language-specific results · MSWE-agent	5.43% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MSWE-agent	#11 of 27 in this displayed comparison; 9.3 pp below the leader. Source notes Source cell: 5.43
DeepSeek-R1 →DeepSeekMulti-SWE-bench · language-specific results · MopenHands	4.65% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MopenHands	#13 of 27 in this displayed comparison; 10.08 pp below the leader. Source notes Source cell: 4.65
Claude-3.5-Sonnet →AnthropicMulti-SWE-bench · language-specific results · MagentLess	3.88% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MagentLess	#14 of 27 in this displayed comparison; 10.85 pp below the leader. Source notes Source cell: 3.88
GPT-4o →OpenAIMulti-SWE-bench · language-specific results · MopenHands	3.88% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MopenHands	#14 of 27 in this displayed comparison; 10.85 pp below the leader. Source notes Source cell: 3.88
OpenAI-o1 →OpenAIMulti-SWE-bench · language-specific results · MSWE-agent	3.88% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MSWE-agent	#14 of 27 in this displayed comparison; 10.85 pp below the leader. Source notes Source cell: 3.88
OpenAI-o1 →OpenAIMulti-SWE-bench · language-specific results · MopenHands	3.88% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MopenHands	#14 of 27 in this displayed comparison; 10.85 pp below the leader. Source notes Source cell: 3.88
Claude-3.7-Sonnet →AnthropicMulti-SWE-bench · language-specific results · MagentLess	3.1% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MagentLess	#18 of 27 in this displayed comparison; 11.63 pp below the leader. Source notes Source cell: 3.10
DeepSeek-R1 →DeepSeekMulti-SWE-bench · language-specific results · MagentLess	3.1% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MagentLess	#18 of 27 in this displayed comparison; 11.63 pp below the leader. Source notes Source cell: 3.10
Doubao-1.5-pro →ByteDanceMulti-SWE-bench · language-specific results · MopenHands	3.1% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MopenHands	#18 of 27 in this displayed comparison; 11.63 pp below the leader. Source notes Source cell: 3.10
GPT-4o →OpenAIMulti-SWE-bench · language-specific results · MSWE-agent	2.33% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MSWE-agent	#21 of 27 in this displayed comparison; 12.4 pp below the leader. Source notes Source cell: 2.33
Qwen2.5-72B-Instruct →QwenMulti-SWE-bench · language-specific results · MopenHands	2.33% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MopenHands	#21 of 27 in this displayed comparison; 12.4 pp below the leader. Source notes Source cell: 2.33
DeepSeek V3 →DeepSeekMulti-SWE-bench · language-specific results · MagentLess	1.55% 1 report	Multi-SWE-bench · Resolved · C++[1]Multi-SWE-bench · language-specific results · MagentLess	#23 of 27 in this displayed comparison; 13.18 pp below the leader. Source notes Source cell: 1.55

Sources

What it measures

Reading the measurement

Published evaluation

Where this fits in an evaluation suite

A reproducible comparison

Continue exploring

AA-3D

Terminal-Bench 4.0

FrontierCode 1.1

Terminal-Bench 2.1