LUNAR - Curated Context: leaderboard
Personalized response quality with only the query-relevant behavioral evidence in context (an oracle setting), averaging GPT-5.1-judged personalization coverage and depth across daily-life domains such as clothing, food, housing and mobility (LUNAR, arXiv 2608.05246v2, Table 4). 1-5 scale; higher is better; 19 models.
Metric: Personalization score (1-5; mean of coverage and depth, GPT-5.1 judge). Source: arxiv.org. Saturation forecast: Around February 2027. 19 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Kimi K2.6 | 4.07 | #113 |
| 2 | Gemini 3 Flash (Preview) | 4.02 | #90 |
| 3 | GLM-5.1 | 3.86 | #109 |
| 4 | DeepSeek V4 Flash | 3.78 | #155 |
| 5 | DeepSeek V4 Pro | 3.66 | #112 |
| 6 | MiniMax-M2.7 | 3.57 | #270 |
| 7 | GPT-4.1 Mini | 3.39 | #363 |
| 8 | Qwen 3 30B A3B 2507 Instruct | 3.37 | #492 |
| 9 | Ling-2.6-1T | 3.35 | #307 |
| 10 | Qwen 3 14B (Non-reasoning) | 3.27 | #542 (Qwen 3 14B) |
| 11 | Ling-2.6-flash | 3.19 | #533 |
| 12 | GLM-4.7 Flash | 3.15 | #543 |
| 13 | Qwen 3 8B (Non-reasoning) | 3.12 | #678 (Qwen 3 8B) |
| 14 | Qwen 3 32B (Non-reasoning) | 3.12 | #449 (Qwen 3 32B) |
| 15 | GPT-4o Mini | 3.08 | #598 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=lunar-curated-context · How It Works · Data refreshed daily, snapshot 2026-11-05.