MacroLens - Real-Estate Rent: leaderboard
Zero-shot residential rent estimation from static property attributes (location, property type, size, rooms, year built, last sale) for 1,000 held-out U.S. addresses, the non-equity transfer task of MacroLens (Trirat et al., 2026). Mean absolute percentage error in percent, per-instance errors capped at 1,000%, for LLMs run with reasoning disabled; lower is better.
Metric: Mean absolute percentage error (%; monthly rent predicted from static property attributes for 1,000 held-out U.S. addresses; per-instance APE capped at 1,000%; zero-shot LLM with reasoning disabled). Source: arxiv.org. Saturation forecast: Estimated already saturated. 5 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-5.1 (Non-reasoning) | 22.82 | #225 (GPT-5.1) |
| 2 | Qwen 3.5 27B (Non-reasoning) | 24.91 | #358 (Qwen 3.5 27B) |
| 3 | Llama 4 Scout | 31.3 | #1787 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=macrolens-real-estate-rent · How It Works · Data refreshed daily, snapshot 2026-11-12.