Tool-augmented agentic reasoning on Evaluation Corpus Low-Budget
26.54Delta Cap Hits (%)Sponge Tool Attack
Evaluation Results
| Method | Links | |||
|---|---|---|---|---|
| Sponge Tool AttackVictim Model=QWEN2-VL-7B2026.01 | 26.54 | — | — | |
| Sponge Tool AttackVictim Model=GPT-4O-MINI2026.01 | 13.35 | — | — | |
| Sponge Tool AttackVictim Model=LLAVA-ONEVISION-7B2026.01 | 12.85 | — | — | |
| Sponge Tool AttackVictim Model=QWEN3-VL-2B2026.01 | 5.3 | — | — | |
| Sponge Tool AttackVictim Model=GPT-4.1-NANO2026.01 | 4.9 | — | — | |
| Sponge Tool AttackVictim Model=Gemma-3-27B2026.01 | 3.83 | — | — |