DeepSWE
DeepSWE - agentic coding benchmark from datacurve/deep-swe dataset evaluating software engineering agent capabilities.
Category
Agentic Coding
Max Score
100.0
Score Type
percent
Active
Yes
Model Scores
Scores for DeepSWE
| Model | Score | Date | Verified |
|---|---|---|---|
| DeepSeek-V4-Flash-0731 |
74.83%
|
31.07.2026 | Verified |
| DeepSeek-V4-Flash |
10.04%
|
13.08.2026 | Verified |
| GLM-5.2 |
63.55%
|
31.07.2026 | Verified |
| Opus-4.8 |
79.78%
|
31.07.2026 | Verified |
| DeepSeek-V4-Pro |
17.61%
|
13.08.2026 | Verified |
| Ornith-1.5-35B-A3B |
30.26%
|
19.08.2026 | Verified |
| Ornith-1.0-35B-A3B |
0.00
|
19.08.2026 | Unverified |
| Qwen3.6-35B-A3B |
0.00
|
19.08.2026 | Unverified |
| Qwen3.5-397B-A17B |
1.38%
|
19.08.2026 | Unverified |
| DeepSeek-V4-Pro-0813 |
86.24%
|
13.08.2026 | Verified |
| Kimi K3 |
92.85%
|
13.08.2026 | Verified |
| Fable-5 |
96.29%
|
13.08.2026 | Verified |
| Hy3 |
38.51%
|
28.08.2026 | Verified |
| Hy4 preview |
88.45%
|
28.08.2026 | Verified |
| Qwen3.8-2.4T-A95B |
77.85%
|
28.08.2026 | Verified |
| GLM-5.3 |
92.02%
|
28.08.2026 | Verified |
| GPT-5.6 Sol |
100.00%
|
28.08.2026 | Verified |
| Claude Opus 5 |
94.64%
|
28.08.2026 | Verified |
| DeepSeek-V4-Flash-Vision-Exp |
81.57%
|
08.09.2026 | Verified |
| DeepSeek V4 Pro |
86.24%
|
28.08.2026 | Verified |