Repository-level code generation on DevEval
442Inference TimeCo-Coder
Evaluation results
| Method | Links | ||||
|---|---|---|---|---|---|
| Co-CoderOrchestration Strategy=Cohesion-based graph-partitioning, Base Model=gpt-5-mini, SDK Version=OpenHands SDK v1.11.42026.05 | 442 | — | 0.18 | 68.1 | |
| Claude Code w/ Agent TeamsOrchestration Strategy=Agent Teams (self-coordinated independent sessions), Base Model=gpt-5-mini, CLI Version=v2.1.119, API Proxy=LiteLLM v1.83.82026.05 | 536 | — | 0.23 | 54.1 | |
| SequentialOrchestration Strategy=Sequential (single coding agent), Base Model=gpt-5-mini, SDK Version=OpenHands SDK v1.11.42026.05 | 800 | — | 0.25 | 56.8 | |
| File-based ParallelOrchestration Strategy=File-based Parallel (leader agent spawns per-file agents), Base Model=gpt-5-mini, SDK Version=OpenHands SDK v1.11.42026.05 | 806 | — | 0.36 | 57.7 |