RecoAtlas: leaderboard
LLM recommendation agents increasingly produce structured recommendation reports: sets of items accompanied by natural-language justifications. Yet existing evaluations often reduce this setting to reranking small shortlisted candidate sets or judge reports mainly by semantic plausibility. We introduce Recommendation Atlas (Agentic Tool-Level Assessment for Shopping), or RecoAtlas, a benchmark and
Metric: Bundle SetHit@20 (self-reported). Source: benchmarklist.com. Saturation forecast: Around February 2027. 12 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Grok 4.1 Fast | 25.2 | #217 |
| 2 | Gemini 2.5 Flash | 24.7 | #245 |
| 3 | GPT-4.1 Mini | 24.4 | #355 |
| 4 | Qwen 3 32B | 24.2 | #443 |
| 5 | Llama 3.3 70B | 22.2 | #579 |
| 6 | DeepSeek R1 Distill Qwen 32B | 22 | #655 |
| 7 | Llama 4 Maverick | 21.4 | #464 |
| 8 | Claude Haiku 4.5 | 21 | #280 |
| 9 | Llama 3.2 11B | 19.6 | #1199 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=recoatlas · How It Works · Data refreshed daily, snapshot 2026-10-20.