InteractiveBench Decision (task pass)
Agents · 2026-10-25
InteractiveBench's Decision category: the 20 Decision tasks, in which the assistant helps make an everyday choice from what it has observed and its own knowledge (for example picking the cheaper of two supplier quotes for a full order). 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 | 85.0 |
| Claude Opus 5 | 70.0 |
| GPT-5.6 Sol | 70.0 |
| GPT-5.5 | 65.0 |
| Muse Spark 1.1 | 55.0 |
| Opus 4.8 | 50.0 |
| Gemini 3.1 Pro Preview | 50.0 |
| GLM-5.3 Flash | 50.0 |
| GPT-5.6 Luna | 40.0 |
| Kimi K3 | 35.0 |