agents benchmark
DeepSWE
DeepSWE is a software engineering agent benchmark evaluated with the mini-swe-agent harness, where each task is solved in an isolated container with no internet access. It measures an agent's ability to autonomously resolve real-world coding issues end to end.
Updated Aug 7, 2026
Published models10
Registry coverage10
MetricScore
EvidenceB
DeepSWE leaderboard
Sorted by the source-provided rank. Higher score is better according to the registry.
| 1 | OP | 72.7% | 100.0% | 10 | C | |
| 2 | OP | 69.6% | 88.9% | 10 | C | |
| 3 | MA | 67.5% | 77.8% | 10 | C | |
| 4 | OP | 67.2% | 66.7% | 10 | C | |
| 5 | DE | 54.4% | 55.6% | 10 | C | |
| 6 | XA | 53.0% | 44.4% | 10 | C | |
| 7 | ZA | 46.2% | 33.3% | 10 | C | |
| 8 | BY | 32.7% | 22.2% | 10 | C | |
| 9 | TE | 28.0% | 11.1% | 10 | C | |
| 10 | BY | 23.0% | 0.0% | 10 | C |
Score distribution
Top published rows on the benchmark's original scale.
DeepSWE highlights
The top published results on this benchmark's own scale.
What is DeepSWE?
Definition and scoring fields from the benchmark registry.
DeepSWE is a software engineering agent benchmark evaluated with the mini-swe-agent harness, where each task is solved in an isolated container with no internet access. It measures an agent's ability to autonomously resolve real-world coding issues end to end.
Scores are shown in ratio. The current registry marks this benchmark as not independently verified with evidence level B.
- Family
- DeepSWE
- Modality
- text
- Primary category
- agents
- Score direction
- higher
- LLMBoard eligible
- No
- Evaluation key
- deepswe|llm-stats-current
Source-native results are preserved. Eligibility for the overall LLMBoard score is a separate policy decision.
FAQ
Common questions about DeepSWE.
Which model scores highest on DeepSWE?
GPT-5.6 Sol is currently ranked first with 72.7%.
What does DeepSWE measure?
DeepSWE is a software engineering agent benchmark evaluated with the mini-swe-agent harness, where each task is solved in an isolated container with no internet access. It measures an agent's ability to autonomously resolve real-world coding issues end to end.
Is a higher score better?
Yes. Higher values rank better for this benchmark.
How many models are compared?
10 unique published model results are currently shown.
Does this benchmark affect the overall score?
This benchmark is preserved as source-native evidence but is not eligible for the current overall score.