InteractiveBench Guidance (task pass)
Agents · 2026-10-25
InteractiveBench's Guidance category: the 20 Guidance tasks, in which the assistant guides the user step by step through a longer job (for example putting away 14 labelled chemical bottles under lab safety rules), tracking progress and correcting mistakes. Same simulator, Gemini 3.1 Pro Human agent and end-state check as the full benchmark; the score is the percentage of these 20 tasks in which every required end state holds.
Top models (higher is better)
| Model | Score |
|---|---|
| Claude Fable 5.1 | 20.0 |
| GPT-5.6 Sol | 20.0 |
| Kimi K3 | 20.0 |
| Muse Spark 1.1 | 20.0 |
| Claude Opus 5 | 15.0 |
| GLM-5.3 Flash | 15.0 |
| Gemini 3.1 Pro Preview | 10.0 |
| GPT-5.5 | 10.0 |
| GPT-5.6 Luna | 5.0 |
| Opus 4.8 | 0.0 |