LUNAR - RAG: leaderboard
Metric: Personalization score (1-5; mean of coverage and depth, GPT-5.1 judge). Source: arxiv.org. 19 models tracked.
Top models
| # | Model | Score |
|---|---|---|
| 1 | Kimi K2.6 | 3.15 |
| 2 | GLM-5.1 | 3.13 |
| 3 | Gemini 3 Flash (Preview) | 3.13 |
| 4 | DeepSeek V4 Flash | 3 |
| 5 | MiniMax-M2.7 | 2.93 |
| 6 | GPT-4.1 Mini | 2.86 |
| 7 | Ling-2.6-1T | 2.79 |
| 8 | DeepSeek V4 Pro | 2.74 |
| 9 | Qwen 3 30B A3B 2507 Instruct | 2.71 |
| 10 | Qwen 3 14B (Non-reasoning) | 2.68 |
| 11 | Ling-2.6-flash | 2.62 |
| 12 | GLM-4.7 Flash | 2.6 |
| 13 | GPT-4o Mini | 2.58 |
| 14 | Qwen 3 8B (Non-reasoning) | 2.58 |
| 15 | Qwen 3 32B (Non-reasoning) | 2.4 |
Interactive version: theaggregate.ai/benchmark?slug=lunar-rag · How It Works · Data refreshed daily, snapshot 2026-09-19.