HarmActionsEval
AI agent safety benchmark measuring how often autonomous LLM agents avoid harmful tool actions under adversarial pressure.
10rows
safe_actions_at_1primary metric
2026-05-06sampled
Metadata
Metrics
SafeActions@1
| Rank | Subject | SafeActions@1 | Model Match | Provenance | Sampled |
|---|---|---|---|---|---|
| 1 | Qwen3.5-397b-a17b | 23.40 |  Qwen3.5 397B A17B qwen-qwen3.5-397b-a17b | Imported | 2026-05-06 |
| 2 | GPT-5.3 | 12.77 | — | Imported | 2026-05-06 |
| 3 | Claude Sonnet 4.6 | 2.84 |  Claude Sonnet 4.6 anthropic-claude-sonnet-4.6 | Imported | 2026-05-06 |
| 4 | Phi 4 Mini Reasoning | 2.84 | — | Imported | 2026-05-06 |
| 5 | Ministral 3 (3B) | 2.13 | — | Imported | 2026-05-06 |
| 6 | GPT-5.4 Mini | 0.71 |  GPT-5.4 Mini openai-gpt-5.4-mini | Imported | 2026-05-06 |
| 7 | Gemini 3.1 Flash Lite | 0.71 | — | Imported | 2026-05-06 |
| 8 | Claude Haiku 4.5 | 0 | Claude Haiku 4.5 anthropic-claude-haiku-4.5 | Imported | 2026-05-06 |
| 9 | Phi 4 Mini Instruct | 0 | — | Imported | 2026-05-06 |
| 10 | Granite 4-H-Tiny | 0 | — | Imported | 2026-05-06 |
No matching rows.