Django Benchmark Suite - Method Generation: leaderboard
A Django benchmark suite (Shibaev et al., 2026) built from a leakage-controlled Django repository snapshot to test whether SWE-bench-oriented training transfers to other repository-level coding tasks. In method generation the model writes a method body from its signature, a synthesized docstring and the rest of the file; pass@1 in percent over 359 held-out methods, a method passing when all its mapped Django tests pass, higher is better.
Metric: pass@1 (%; Method generation: the model writes a method body from its signature, a synthesized docstring and the rest of its source file; the output is inserted into the file and counts only when every Django test mapped to the method passes; 359 held-out methods of a leakage-controlled Django repository snapshot; greedy decoding, no agent scaffold). Source: arxiv.org. Saturation forecast: Estimated already saturated. 18 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.5 | 88.02 | #156 |
| 2 | Gemini 3 Pro | 86.35 | #137 |
| 3 | Gemini 3 Flash | 85.79 | #164 |
| 4 | GPT-5 | 84.68 | #167 |
| 5 | Gemini 2.5 Pro | 83.01 | #266 |
| 6 | Claude Sonnet 4.5 | 82.73 | #250 |
| 7 | GPT-4.1 | 76.6 | #495 |
| 8 | Claude Haiku 4.5 | 74.65 | #568 |
| 9 | Qwen 3 32B | 67.41 | #960 |
| 10 | Qwen 2.5 Coder 32B Instruct | 64.07 | #1562 |
| 11 | Qwen 2.5 Coder 7B Instruct | 52.92 | #3852 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=django-benchmark-suite-method-generation · How It Works · Data refreshed daily, snapshot 2026-11-13.