VIVID: leaderboard
VIVID (Vietnamese Idioms for Validation and Interpretation Depth, 2026) tests culturally grounded understanding of 1,636 Vietnamese idioms and proverbs. Models explain each idiom and a GPT-4.1 judge rates the explanation from 0 to 5 under zero-shot prompting; higher is better.
Metric: Explanation score, 0 to 5 (out of 5); zero-shot explanation of each of the 1,636 Vietnamese idioms and proverbs, rated 0-5 for semantic accuracy, completeness and native-speaker alignment by a GPT-4.1 judge with the aspect-based rubric; item-weighted mean over seven topics). Source: arxiv.org. Saturation forecast: Estimated already saturated. 8 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | GPT-4o | 2.46 | #741 |
| 2 | Gemini 2.5 Flash | 2.29 | #452 |
| 3 | Qwen 3 14B | 1.09 | #1235 |
| 4 | Llama 4 Scout Base | 0.91 | #1890 |
| 5 | Llama-SEA-LION-v3-8B-IT | 0.58 | #2317 |
No result here: #1 Claude Fable 5.2, #2 Gemini 4 Carbon, #5 Claude Opus 5.5.
Interactive version: theaggregate.ai/benchmark?slug=vivid · How It Works · Data refreshed daily, snapshot 2026-11-12.