Mistral Nemo 12B Instruct

Overview

Mistral AI and NVIDIA's co-developed 12B instruct model with 128k context, the Tekken tokenizer, and strong European multilingual coverage.

When to pick this model

• Multilingual chat across European languages
• Long-context summarization and RAG
• Replacing Mistral 7B with a noticeable quality bump
• Apache 2.0 commercial deployments on a single 24GB GPU
• NVIDIA-tuned inference stacks

VRAM requirements by quantization

Quantization	VRAM required
Q4_K_M (recommended)	7 GB
Q5_K_M	9 GB
Q8_0	13 GB
FP16 (no quantization)	24 GB

VRAM figures include model weights plus a typical 8k KV cache and ~600 MB runtime overhead (Ollama / llama.cpp baseline). Add headroom for higher context lengths.

Published benchmark scores

Benchmark	Score
MMLU	68
HellaSwag	83.5
Winogrande	76.8

Scores published by the model author or aggregated from public leaderboards. Re-measured monthly by our editorial team.

Strengths

• 128k context window
• Strong European multilingual performance
• Apache 2.0 license
• Efficient Tekken tokenizer reduces token counts

Limitations

• Reasoning trails Mistral Small 3.1
• No vision
• Eclipsed by Small 3 on most general benchmarks

Architecture & training

Architecture: Dense Transformer · GQA · Tekken tokenizer (131k vocab)

Training: Co-trained by Mistral × NVIDIA. European multilingual corpus.

Quick start

ollama run mistral-nemo:12b

Or use the open-source MCP server to query this model from Claude Desktop, Cursor, or any MCP-compatible client.