Shopping Companion Bench: leaderboard
Metric: Success rate (%): share of tasks whose recommendation has the right number of products, meets the request, matches the user's long-term preferences and fits the budget, averaged over both task types, on Shopping Companion Bench's 200 test tasks (100 single-product and 100 add-on-deal recommendations), zero-shot: the agent searches about 50 sessions (about 106K tokens) of a user's long-term conversation memory for the needed preferences and then searches a 1.3M-product Lazada catalog; a GPT-5 judge checks the recommendation; higher is better. Source: arxiv.org. Saturation forecast: Around January 2027. 10 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 64.5 | #91 |
| 2 | GPT-4.1 | 51 | #240 |
| 3 | GPT-4o | 49 | #333 |
| 4 | Qwen 3 Max | 48 | #201 |
| 5 | Kimi K2 | 44.5 | #236 |
| 6 | DeepSeek R1 | 40.5 | #245 |
| 7 | Qwen 3 Next 80B A3B | 37.5 | #306 |
| 8 | Qwen 3 30B A3B | 33 | #488 |
| 9 | Gemma 3 27B | 30.5 | #596 |
| 10 | Qwen 3 4B | 25 | #823 |
No result here: #3 Claude Opus 5.5, #5 GPT-6 Astra, #8 Claude Fable 5.1.
Interactive version: theaggregate.ai/benchmark?slug=shopping-companion-bench · How It Works · Data refreshed daily, snapshot 2026-10-11.