File-level Localization on LocBench
77.1Top-1 AccuracyLARGERTuned
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| LARGERTunedCategory=Agent-based (graph-enhanced), Backbone=GPT-5.2, Setting=Tuned2026.05 | 77.1 | 89.1 | 81.5 | 92.3 | |
| LARGERFixedCategory=Agent-based (graph-enhanced), Backbone=GPT-5.2, Setting=Fixed2026.05 | 72.9 | 87 | 77.4 | 90.1 | |
| Claude Code*Category=Agent-based, Backbone=Claude-Opus-4.62026.05 | 65.1 | 75.2 | 71.8 | 85.6 | |
| CodexCategory=Agent-based, Backbone=GPT-5.22026.05 | 63.2 | 74.1 | 69.3 | 82.8 | |
| mini-swe-agentCategory=Agent-based, Backbone=GPT-5.22026.05 | 60.6 | 70.7 | 66.9 | 79.7 | |
| SWE-agentCategory=Agent-based, Backbone=GPT-5.22026.05 | 60 | 70.9 | 65.8 | 79.3 | |
| OpenHandsCategory=Agent-based, Backbone=GPT-5.22026.05 | 57.7 | 67.5 | 63.2 | 75.3 | |
| AgentlessCategory=Procedure-based, Backbone=GPT-5.22026.05 | 56.1 | 68.9 | 61.2 | 76.6 | |
| LocAgentCategory=Agent-based (graph-enhanced), Backbone=GPT-5.22026.05 | 56 | 65.3 | 62 | 73.8 | |
| CoSILCategory=Agent-based (graph-enhanced), Backbone=GPT-5.22026.05 | 54.6 | 68.2 | 59.7 | 75.7 | |
| BM25Category=Lexical retrieval2026.05 | 23.4 | 49.3 | 26.9 | 55.7 |