Mistral NeMo 12B
Mistral NeMo, a 12B model built with Nvidia, ships 128K context, Apache 2.0 and the multilingual Tekken tokenizer.
Drop-in successor to Mistral 7B with a 128K window. The Tekken tokenizer (100+ languages) is about 30% more efficient on code and several major languages and 2-3x better on Korean and Arabic.
- Date
- Thursday, 18 July 2024
- Lab
- Mistral AI
- Kind
- open-weights
- Access
- open weights
Figures
| Measure | Value | Measured by |
|---|---|---|
| Context window | 128K tokens 12B params, Apache 2.0 | company |
Sources
This record was checked against its sources on 6 October 2026. How we check