Software Engineering Issue Solving on SWE-Bench Verified
54AccuracyOpenThinkerAgent-32B
Evaluation results
| Method | Links | ||
|---|---|---|---|
| OpenThinkerAgent-32BTrain Size=100K, Method=SFT, Backbone=Qwen3-32B2026.06 | 54 | — | |
| SWE-Lego-Qwen3-32BTrain Size=18K, Method=SFT, Backbone=Qwen3-32B2026.06 | 51 | — | |
| SERA-32BTrain Size=25K, Method=SFT, Backbone=Qwen3-32B2026.06 | 49.4 | — | |
| PromptBridgeModel Role=Target Model, Model=o3, Source Model=o4-mini, Prompt Type=Model-adaptive prompt transfer2025.12 | 46 | 27.39 | |
| OptimizedModel Role=Source Model, Model=o4-mini, Prompt Type=Optimized prompt for source model2025.12 | 42.2 | — | |
| DeepSWE-PreviewTrain Size=4.5K, Method=RL, Backbone=Qwen3-32B2026.06 | 42.2 | — | |
| Nemotron-Terminal-32BTrain Size=264K, Method=SFT, Backbone=Qwen3-32B2026.06 | 41.9 | — | |
| SA-SWE-32BTrain Size=4.5K, Method=RL, Backbone=Qwen3-32B2026.06 | 39.4 | — | |
| OriginalModel Role=Source Model, Model=o4-mini, Prompt Type=Default template from mini-SWE-agent2025.12 | 38.6 | — | |
| Direct TransferModel Role=Target Model, Model=o3, Source Model=o4-mini, Prompt Type=Optimized prompt from source model2025.12 | 33.4 | 0 | |
| OriginalModel Role=Target Model, Model=o3, Prompt Type=Default template from mini-SWE-agent2025.12 | 32 | — | |
| Qwen3-32BTrain Size=N/A, Method=N/A, Backbone=Qwen3-32B2026.06 | 29.1 | — | |
| PromptBridgeModel Role=Target Model, Model=Llama-3.1-70B-Instruct, Source Model=o4-mini, Prompt Type=Model-adaptive prompt transfer2025.12 | 8.8 | 15.79 | |
| Direct TransferModel Role=Target Model, Model=Llama-3.1-70B-Instruct, Source Model=o4-mini, Prompt Type=Optimized prompt from source model2025.12 | 7.6 | 0 | |
| OriginalModel Role=Target Model, Model=Llama-3.1-70B-Instruct, Prompt Type=Default template from mini-SWE-agent2025.12 | 6.6 | — |