Ministral 3 3B
Ministral 3 3B is an open-weights language model from Mistral, released December 2, 2025. It has 3.9B parameters (dense). The context window is 128K tokens. API price: $0.1 input / $0.1 output per million tokens.
Today, we announce Mistral 3, the next generation of Mistral models. Mistral 3 includes three state-of-the-art small, dense models (14B, 8B, and 3B) and Mistral Large 3 – our most capable model to date – a sparse mixture-of-experts trained with 41B active and 675B total parameters.
Check it against your own PC →
Can you run Ministral 3 3B locally?
The smallest common GPU it runs on is 6 GB of VRAM (with 32 GB of RAM): Q5_K_M, fully on GPU, roughly 40 tokens per second. Estimates assume 32 GB of DDR5 RAM, a 16K context and llama.cpp-style GGUF quantization. Enter your exact hardware in the live app.
| GPU memory | Verdict | Best quant | VRAM | RAM | tok/s |
|---|---|---|---|---|---|
| 6 GB VRAM | Runs on GPU | Q5_K_M (excellent) | 5.1 GB | – | ~40 |
| 8 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.5 GB | – | ~49 |
| 12 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.5 GB | – | ~64 |
| 16 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.5 GB | – | ~87 |
| 24 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.5 GB | – | ~150 |
| 32 GB VRAM | Runs on GPU | Q6_K (≈ lossless) | 5.5 GB | – | ~250 |
Ready-made quantizations on Hugging Face: Mungert/Ministral-3-3B-Reasoning-2512-GGUF · mradermacher/Ministral-3-3B-Base-2512-GGUF · mradermacher/Ministral-3-3B-Base-2512-i1-GGUF · Bercraft/Voxtral-Mini-4B-Realtime-2602-GGUF · praiselab-picuslab/Voxtral-Mini-4B-Realtime-2602-GGUF · mlx-community/Ministral-3-3B-Instruct-2512-4bit
Benchmark results
Epoch AI has not published results for Ministral 3 3B yet. New models are usually evaluated within days to weeks.
Links
Announcement · Hugging Face: mistralai/Ministral-3-3B-Base-2512
More from Mistral
Mistral Large 4 · Mistral Large 4 Preview · Mistral Large 4 1T A52B · Leanstral 1.5 · Mistral Medium 3.5 · Mistral Small 4 · Devstral Small 2 · Devstral 2