Agentic Execution Trace Analysis on TraceBench Deep Research BrowseComp
81Overall ScoreTraceSIR
Evaluation Results
| Method | Links | |||||||
|---|---|---|---|---|---|---|---|---|
| TraceSIRBackbone=Claude-4.62026.02 | 81 | 8.5 | 8.5 | 8 | 7 | 8.5 | 1 | |
| ClaudeCodeBackbone=Claude-4.62026.02 | 66 | 6.5 | 7.5 | 5.5 | 7 | 6.5 | 2 | |
| TraceSIRBackbone=GLM-52026.02 | 65 | 7.5 | 6.5 | 5.5 | 6 | 7 | 3 | |
| ClaudeCodeBackbone=GLM-52026.02 | 55 | 5.5 | 4.5 | 5.5 | 6 | 6 | 4 |