ComboShoppingBench - Response Quality: leaderboard
Metric: Pass rate (%; LLM-judged). Source: arxiv.org. 22 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Claude Opus 4.8 (Thinking) | 96.6 |
| 2 | Gemini 3.1 Pro (Preview) (Thinking) | 93.8 |
| 3 | GPT-5.5 (Non-reasoning) | 93.1 |
| 4 | Kimi K2.6 (Thinking) | 92.8 |
| 5 | GPT-5.5 | 92.4 |
| 6 | Claude Opus 4.6 (Thinking) | 92.1 |
| 7 | GLM-5.2 (Non-reasoning) | 90.7 |
| 8 | Seed 2.0 Pro (Thinking) | 81.8 |
| 9 | Qwen 3.6 27B (Thinking) | 78.4 |
| 10 | Kimi K2.6 (Non-reasoning) | 75.6 |
| 11 | MiniMax M3 (Thinking) | 72.5 |
| 12 | Claude Sonnet 4.6 (Non-reasoning) | 68.4 |
| 13 | Claude Sonnet 4.6 (Thinking) | 59.8 |
| 14 | Qwen 3.6 27B (Non-reasoning) | 51.2 |
| 15 | Claude Opus 4.6 (Non-reasoning) | 19.6 |
Interactive version: theaggregate.ai/benchmark?slug=comboshoppingbench-response-quality · How It Works · Data refreshed daily, snapshot 2026-09-19.