Hebrew LLM - Nikud: leaderboard
The Nikud task of the Hebrew LLM Leaderboard — adding nikud (Hebrew vowel diacritics) to unvocalized Hebrew text; score (%).
Metric: Score (%). Source: huggingface.co. Saturation forecast: Around January 2027. 34 models tracked.
Top models
| # | Model | Score | Overall rank |
|---|---|---|---|
| 1 | Gemini 3 Flash (Preview) | 88.37 | #82 |
| 2 | Gemini 3 Pro (Preview) | 87.58 | #66 |
| 3 | Gemini 2.5 Pro | 87.15 | #152 |
| 4 | GPT-5.1 | 80.55 | #135 |
| 5 | Gemini 2.5 Flash | 79.52 | #245 |
| 6 | Qwen 3.5 397B A17B | 78.23 | #146 |
| 7 | DeepSeek V3.1 | 75.29 | #264 |
| 8 | GPT-4o | 74.82 | #343 |
| 9 | Gemma 4 31B (IT) | 73.41 | #176 |
| 10 | Gemini 2.5 Flash Lite | 70.73 | #422 |
| 11 | Gemma 4 26B A4B (IT) | 68.07 | #297 |
| 12 | Qwen 3.5 122B A10B | 67.11 | #173 |
| 13 | GPT-5 Mini | 66.22 | #186 |
| 14 | Claude Sonnet 4.5 | 62.95 | #142 |
| 15 | Claude Haiku 4.5 | 62.53 | #280 |
No result here: #3 Claude Opus 5.5, #6 GPT-6 Astra, #7 GPT-6.1 Sol.
Interactive version: theaggregate.ai/benchmark?slug=hebrew-llm-nikud · How It Works · Data refreshed daily, snapshot 2026-10-20.