Trajectory Attribution on Who&When Algorithm-Generated n=126
68Agent-level AccuracyFALAT
Evaluation Results
| Method | Links | ||
|---|---|---|---|
| FALATBackbone=MiniMax 2.52026.05 | 68 | 46 | |
| FALATBackbone=GPT-5.4-Mini2026.05 | 66.5 | 42 | |
| DirectBackbone=DeepSeek-R12026.05 | 65.1 | 29.5 | |
| AgentTracerBackbone=AgentTracer-8B2026.05 | 63.7 | 37.3 | |
| DirectBackbone=Qwen3-8B2026.05 | 60.3 | 5.6 | |
| FALATBackbone=DeepSeek V3.22026.05 | 59.7 | 38 | |
| DirectBackbone=GPT-4.12026.05 | 59.5 | 21.9 | |
| CHIEFBackbone=GPT-5.4-Mini2026.05 | 59 | 27 | |
| DirectBackbone=Qwen3-32B2026.05 | 57.9 | 8.7 | |
| DirectBackbone=Gemini-2.5-Pro2026.05 | 57.1 | 25.9 | |
| CHIEFBackbone=MiniMax 2.52026.05 | 54.8 | 37.3 | |
| CHIEFBackbone=DeepSeek V3.22026.05 | 52 | 30 | |
| DirectBackbone=Claude-Sonnet-42026.05 | 51.1 | 38.8 | |
| DirectBackbone=LLaMA-3.2-3B2026.05 | 45.2 | 8.7 | |
| DirectBackbone=Qwen3-Coder2026.05 | 36.5 | 32.5 |