Tool-augmented agentic reasoning on Evaluation Corpus High-Budget
24.23Delta Cap Hits (%)Sponge Tool Attack
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Sponge Tool AttackVictim Model=QWEN2-VL-7B2026.01 | 24.23 | — | — | |
| Sponge Tool AttackVictim Model=LLAVA-ONEVISION-7B2026.01 | 11.04 | — | — | |
| Sponge Tool AttackVictim Model=GPT-4O-MINI2026.01 | 8.9 | — | — | |
| Sponge Tool AttackVictim Model=GPT-4.1-NANO2026.01 | 3.89 | — | — | |
| Sponge Tool AttackVictim Model=QWEN3-VL-2B2026.01 | 3.66 | — | — | |
| Sponge Tool AttackVictim Model=Gemma-3-27B2026.01 | 2.37 | — | — |