B2V-Bench (Label Names Only): leaderboard
B2V-Bench (Peking University and Alibaba, 2026) asks language models to identify the consumer values, such as brand trust, perceived risk or price sensitivity, behind a real Taobao purchase decision from a summary of the shopper behaviour before the purchase. Macro-F1 in percent over 28 value labels plus none on 544 episodes labelled by psychology experts, with a prompt that lists only the label names; higher is better.
Metric: Macro-F1 (%; multi-label classification of the consumer values, 28 labels of the E-commerce Consumption Value Taxonomy plus none, expressed in each of 544 Taobao purchase-decision episodes labelled from scratch by psychology experts; the model reads the GPT-5.4-written summary of the pre-purchase behaviour log; prompt lists the value label names only). Source: arxiv.org. Saturation forecast: Around 2029. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5 | 31.4 | #163 |
| 2 | GLM-5 | 30.8 | #256 |
| 3 | Kimi K2.6 | 27.5 | #173 |
| 4 | DeepSeek V3.2 | 26.6 | #403 |
| 5 | Claude Haiku 4.5 | 25.6 | #568 |
| 6 | Qwen 3 235B A22B | 25.3 | #622 |
| 7 | GPT-4o Mini | 24.8 | #1627 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=b2v-bench-label-names-only · How It Works · Data refreshed daily, snapshot 2026-11-12.