PAUSE - Shopping: leaderboard
PAUSE (KDD 2026) evaluates personal AI assistants in a simulated service environment of user-owned data, logs and accounts, with tools gated by user permissions and system configuration and a simulated user. This board covers 60 shopping tasks verified from the final environment state (product, quantity and size, voucher and budget); the score is the aggregated task score in percent, higher is better.
Metric: Task score (%; 60 constrained shopping tasks, state-based check of the purchased product, quantity and size, voucher use and budget, aggregated per task; multi-turn tasks in a simulated personal service environment with 50 assistant tools, user-controlled permissions and system configurations and a Gemini-3-Flash user simulator; higher is better). Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Pro | 72.1 | #137 |
| 2 | GPT-5 | 69.1 | #167 |
| 3 | Gemini 3 Flash | 59 | #164 |
| 4 | DeepSeek V3.2 (Thinking) | 55 | #406 (DeepSeek V3.2) |
| 5 | GPT-5 Mini | 47.3 | #348 |
| 6 | DeepSeek V3.2 (Non-reasoning) | 41.7 | #406 (DeepSeek V3.2) |
| 7 | Gemini 2.5 Pro | 37.7 | #266 |
| 8 | GPT-4.1 | 19.7 | #495 |
| 9 | GPT-4.1 Mini | 18.3 | #779 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=pause-shopping · How It Works · Data refreshed daily, snapshot 2026-11-13.