Shopping Reasoning Bench - Multi-Turn: leaderboard
Weighted rubric pass rate on the 293 multi-turn missions of Shopping Reasoning Bench (Amazon, 2026), averaged over conversation turns, where preferences and constraints evolve across the dialogue. Percent; higher is better.
Metric: Weighted pass rate (%) on the 293 multi-turn expert-authored shopping missions scored by importance-weighted binary rubrics written by retail domain experts (weighted pass rate); each model answers with its native web search tool at default inference settings and a Claude Sonnet 4.5 judge (temperature 0) grades every rubric, averaged over turns; higher is better. Source: arxiv.org. Saturation forecast: Around February 2027. 9 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Opus 4.7 | 78.5 | #79 |
| 2 | Gemini 3.1 Pro (Preview) | 77.7 | #94 |
| 3 | Gemini 3 Flash | 75.7 | #164 |
| 4 | Gemini 3.1 Flash Lite | 73.5 | #364 |
| 5 | Claude Sonnet 4.5 | 71.3 | #250 |
| 6 | GPT-5.4 | 71 | #123 |
| 7 | GPT-5.4 Mini | 65.8 | #389 |
| 8 | GPT-5.4 Nano | 61.9 | #635 |
| 9 | Claude Haiku 4.5 | 59.1 | #568 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=shopping-reasoning-bench-multi-turn · How It Works · Data refreshed daily, snapshot 2026-11-13.