Robustness Evaluation on Robustness benchmark
100Indirect PI AccuracyShellGames
Evaluation Results
| Method | Links | ||||
|---|---|---|---|---|---|
| ShellGames2026.06 | 100 | 93.8 | 95 | 95 | |
| E-ShelLM2026.06 | 75 | 18.8 | 80 | 55 | |
| GPT-5.4 Mini2026.06 | 50 | 0 | 85 | 47.5 | |
| ShelLM2026.06 | 50 | 18.8 | 90 | 57.5 |
| Method | Links | ||||
|---|---|---|---|---|---|
| ShellGames2026.06 | 100 | 93.8 | 95 | 95 | |
| E-ShelLM2026.06 | 75 | 18.8 | 80 | 55 | |
| GPT-5.4 Mini2026.06 | 50 | 0 | 85 | 47.5 | |
| ShelLM2026.06 | 50 | 18.8 | 90 | 57.5 |