ComboShoppingBench - Rule-Based Validation: leaderboard
Metric: Pass rate (%; all four rule-based checks). Source: arxiv.org. 22 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | GPT-5.5 | 83.8 |
| 2 | Gemini 3.1 Pro (Preview) (Thinking) | 82.8 |
| 3 | Claude Opus 4.6 (Non-reasoning) | 75.9 |
| 4 | Claude Opus 4.6 (Thinking) | 73.9 |
| 5 | Claude Opus 4.8 (Thinking) | 73.5 |
| 6 | Kimi K2.6 (Thinking) | 67.4 |
| 7 | Claude Sonnet 4.6 (Thinking) | 65.6 |
| 8 | Claude Sonnet 4.6 (Non-reasoning) | 65.3 |
| 9 | GLM-5.2 (Non-reasoning) | 57 |
| 10 | MiniMax M3 (Thinking) | 54 |
| 11 | Qwen 3.6 27B (Non-reasoning) | 49.1 |
| 12 | DeepSeek V4 Pro (Non-reasoning) | 48.8 |
| 13 | Seed 2.0 Pro (Thinking) | 46.4 |
| 14 | Kimi K2.6 (Non-reasoning) | 35.7 |
| 15 | GPT-5.5 (Non-reasoning) | 28.5 |
Interactive version: theaggregate.ai/benchmark?slug=comboshoppingbench-rule-based-validation · How It Works · Data refreshed daily, snapshot 2026-09-19.