Vibe-Bench - Personalized Fine-tuning - Response BLEU-4: leaderboard
Vibe-Bench tests personalization from user histories: selecting an emotion-regulation strategy, generating a personalized response, and predicting occupational preferences. The paper separates no-history, full-history, profile, retrieval and personalized fine-tuning protocols. P-SFT checkpoints retain distinct identities from their untrained backbones. This series reports Vibe-Bench - Personalized Fine-tuning - Response BLEU-4 from 2609.00921v1, A11.T10: Score (0-100). Higher is better.
Metric: Score (0-100). Source: arxiv.org. Saturation forecast: Rough model projection: around 2026. 6 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Qwen3-8B (Vibe-Bench P-SFT) | 11.36 | |
| 2 | Llama-3.2-3B (Vibe-Bench P-SFT) | 10.8 | |
| 3 | Llama-3.1-8B (Vibe-Bench P-SFT) | 10.64 | |
| 4 | Qwen3-14B (Vibe-Bench P-SFT) | 10.37 | |
| 5 | Qwen3-4B (Vibe-Bench P-SFT) | 10.22 | |
| 6 | Ministral-8B (Vibe-Bench P-SFT) | 1.24 |
Interactive version: theaggregate.ai/benchmark?slug=vibe-bench-personalized-fine-tuning-response-bleu-4 · How It Works · Data refreshed daily, snapshot 2026-11-08.