AgentHarm Verified (benign) (FRR)Browse 284 families
AgentHarm Verified (benign) (FRR)
Can the model plan, use tools, and finish a multi-step task?
Version / measure
- Domain
- Agents
- Ranked models
- 4
- Reported values
- 4
- Reports
- 1 · lab
- Direction
- Lower is better
- Status
- Version not specified
Rankings
Best reported score per model across effort settings
| # | Model | Best setting | Score |
|---|---|---|---|
| 1 | Gemini 3.1 ProGoogle | highcost n/a1 value · Jul 9 · source | 5.7% |
| 2 | Muse Spark 1.1Meta | xhighcost n/a1 value · Jul 9 · source | 16.5% |
| 3 | GPT-5.5OpenAI | xhighcost n/a1 value · Jul 9 · source | 21.6% |
| 4 | Claude Opus 4.8Anthropic | maxcost n/a1 value · Jul 9 · source | 37.5% |
Effort curve
Score vs. reported cost. Lines connect complete sweeps only.
No cost-linked effort sweep for this version.
Definition and comparison boundarypublic methodology
Can the model plan, use tools, and finish a multi-step task?
Agentic execution across tools, environments, or long-running workflows. Lower is better. The source reports a false-refusal rate as a percentage.
Harness, tools, prompts, attempt count, and benchmark version can materially change the result. Compare within one reporting context.
Method / source ↗