MMShopBench: leaderboard
Metric: Judge success at 1 (%; share of cases in which the first recommended product satisfies the annotated purchase intent and every mandatory requirement according to a GPT-5.5 multimodal judge given the product evidence; multimodal multi-turn shopping requests from real assistant logs, answered by a tool-using agent over a frozen offline catalog with text and region-aware image retrieval (at most eight tool steps), followed by evidence-grounded verification and selection of the candidate pool by the policy model). Source: arxiv.org. Saturation forecast: Around December 2026. 5 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Gemini 3.1 Pro (Preview) | 64.7 |
| 2 | Claude Opus 4.8 (Thinking) | 64.4 |
| 3 | MiniMax-M2.7 | 20.1 |
| 4 | Qwen 3.5 9B (Non-reasoning) | 15.9 |
| 5 | Qwen 3.5 122B A10B (Non-reasoning) | 5.5 |
Interactive version: theaggregate.ai/benchmark?slug=mmshopbench · How It Works · Data refreshed daily, snapshot 2026-09-29.