RealWorldShop: leaderboard
LLMs as conversational shopping agents over full sessions against a real product catalog: about 1,200 shopping scenarios with an LLM-simulated customer who reveals and revises needs gradually. Scored by session success rate, the share of sessions that end in a grounded, actionable purchase decision within 30 assistant turns; higher is better (RealWorldShop, 2026).
Metric: Session success rate (%): share of simulated shopping sessions that end in a grounded, actionable purchase decision (Kimi-K2.6 user simulator, at most 30 assistant turns); higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 7 models tracked. Updated 2 Oct 2026.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 71 | #177 |
| 2 | Kimi K2.6 | 67 | #191 |
| 3 | DeepSeek V3.2 | 56 | #417 |
| 4 | Qwen 3.5 27B | 55 | #374 |
| 5 | Qwen 3.5 35B A3B | 53 | #490 |
| 6 | GPT-4o | 38 | #754 |
| 7 | Gemini 2.5 Flash | 38 | #467 |
No result here: #1 Claude Fable 5.2, #2 GPT-6.2 Astra, #3 Gemini 4 Carbon.
Interactive version: theaggregate.ai/benchmark?slug=realworldshop · How It Works · Data refreshed daily, snapshot 2026-11-17.