EComStage - Intent Recognition: leaderboard
Evaluates text-based e-commerce assistance across seven perception, planning and action tasks. The overall score and task results share the main-table evaluation protocol. This series reports Intent Recognition using Task Score (%). Source: the original paper result table.
Metric: Task Score (%). Source: arxiv.org. Saturation forecast: Estimated already saturated. 33 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Claude Sonnet 4 | 90.42 | #207 |
| 2 | GPT-4o | 90.07 | #357 |
| 3 | Qwen 2.5 72B Instruct | 89.83 | #469 |
| 4 | DeepSeek R1 | 89.61 | #262 |
| 5 | Gemini 2.5 Pro | 89.03 | #159 |
| 6 | GPT-OSS-120B | 88.59 | #346 |
| 7 | DeepSeek V3 | 88.59 | #329 |
| 8 | Qwen 2.5 32B Instruct | 88.15 | #509 |
| 9 | Qwen 3 235B A22B Instruct | 87.78 | #336 |
| 10 | Qwen 3 32B | 87.42 | #449 |
| 11 | Qwen 3 14B | 86.98 | #542 |
| 12 | Qwen 3 235B A22B | 86.91 | #323 |
| 13 | Llama 3.3 70B | 86.83 | #578 |
| 14 | InternVL3-14B | 86.76 | #506 |
| 15 | Qwen 3 30B A3B Instruct | 86.32 | #487 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=ecomstage-intent-recognition · How It Works · Data refreshed daily, snapshot 2026-11-05.