ComboShoppingBench - Overall Success: leaderboard
ComboShoppingBench (arXiv 2608.09282v1, Table 2): 22 tool-using agent configurations, 11 models each with thinking enabled and disabled, on 291 compositional shopping tasks over a frozen catalog with coupons and budgets. Share of tasks passing every LLM-judged and rule-based check; percent, higher is better.
Metric: Pass rate (%; all judged and rule-based checks). Source: arxiv.org. Saturation forecast: Around February 2027. 22 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.5 | 61.2 | #33 |
| 2 | Gemini 3.1 Pro (Preview) (Thinking) | 50.2 | #64 (Gemini 3.1 Pro (Preview)) |
| 3 | Claude Opus 4.6 (Thinking) | 49.8 | #67 (Claude Opus 4.6) |
| 4 | Claude Opus 4.8 (Thinking) | 49.5 | #39 (Claude Opus 4.8) |
| 5 | Kimi K2.6 (Thinking) | 44 | #113 (Kimi K2.6) |
| 6 | GLM-5.2 (Non-reasoning) | 30.2 | #80 (GLM-5.2) |
| 7 | Claude Sonnet 4.6 (Thinking) | 27.8 | #94 (Claude Sonnet 4.6) |
| 8 | MiniMax-M3 (Thinking) | 25.4 | #151 (MiniMax-M3) |
| 9 | Claude Sonnet 4.6 (Non-reasoning) | 23.4 | #94 (Claude Sonnet 4.6) |
| 10 | Claude Opus 4.8 (Non-reasoning) | 21.3 | #39 (Claude Opus 4.8) |
| 11 | GPT-5.5 (Non-reasoning) | 14.4 | #33 (GPT-5.5) |
| 12 | Kimi K2.6 (Non-reasoning) | 14.1 | #113 (Kimi K2.6) |
| 13 | Qwen 3.6 27B (Non-reasoning) | 12.4 | #182 (Qwen 3.6 27B) |
| 14 | Claude Opus 4.6 (Non-reasoning) | 11.3 | #67 (Claude Opus 4.6) |
| 15 | Qwen 3.6 27B (Thinking) | 5.2 | #182 (Qwen 3.6 27B) |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=comboshoppingbench-overall-success · How It Works · Data refreshed daily, snapshot 2026-11-05.