How much VRAM does Ministral 3 14B need?
Ministral 3 14B needs 7.9 GB for its weights at 4-bit, 10.0 GB with an 8K context and 13.7 GB with 32K. At 8-bit, 15.9 GB.
- Parameters
- 13.9B
- Longest context
- 256K
- Released
- Dec 2025
- License
- Apache 2.0
- Strongest at
- Coding, Agents, Docs
Memory at each precision
| Precision | Weights | With 8K | With 32K | With 128K |
|---|---|---|---|---|
| As released (FP8) | 14.6 GB | 16.8 GB | 20.5 GB | 35.5 GB |
| 8-bit | 13.8 GB | 15.9 GB | 19.7 GB | 34.7 GB |
| 6-bit | 10.7 GB | 12.8 GB | 16.5 GB | 31.5 GB |
| 5-bit | 9.2 GB | 11.3 GB | 15.1 GB | 30.1 GB |
| 4-bit | 7.9 GB | 10.0 GB | 13.7 GB | 28.7 GB |
| 3-bit | 6.3 GB | 8.5 GB | 12.2 GB | 27.2 GB |
| 2-bit | 4.9 GB | 7.0 GB | 10.7 GB | 25.7 GB |
| 1.6-bit | 2.7 GB | 4.8 GB | 8.5 GB | 23.5 GB |
One conversation, 16-bit context cache, runtime overhead included.
What runs it
| Machine | Best fit | Context | Speed | First reply |
|---|---|---|---|---|
| RTX 3060 12 GB | 4-bit | 16K | 30 tok/s | 2.2 s |
| RTX 4060 Ti 16 GB | 6-bit | 20K | 18 tok/s | 1.3 s |
| RTX 3090 | 8-bit | 48K | 44 tok/s | 0.8 s |
| RTX 4090 | 8-bit | 48K | 47 tok/s | 0.3 s |
| RTX 5090 | 8-bit | 96K | 80 tok/s | 0.3 s |
| 2× RTX 3090 | 8-bit | 192K | 44 tok/s | 0.5 s |
| RTX PRO 6000 Blackwell | 8-bit | 256K | 80 tok/s | 0.2 s |
| M4 Max 128 GB | 8-bit | 256K | 23 tok/s | 4.5 s |
| M5 Max 128 GB | 8-bit | 256K | 26 tok/s | 1.3 s |
| Ryzen AI Max+ 395 128 GB | 8-bit | 256K | 11 tok/s | 2.6 s |
| DGX Spark 128 GB | 8-bit | 256K | 13 tok/s | 0.9 s |
| M3 Ultra 256 GB | 8-bit | 256K | 34 tok/s | 2.5 s |
| M5 Ultra 512 GB | 8-bit | 256K | 47 tok/s | 0.8 s |
| H100 SXM | 8-bit | 256K | 139 tok/s | 0.2 s |
| DGX Station GB300 | 8-bit | 256K | 248 tok/s | 0.1 s |
| 8× H200 NVL | 8-bit | 256K | 185 tok/s | 0.1 s |
Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.
How to run it
- Ollama
ollama run ministral-3:14b- llama.cpp
llama-server -hf unsloth/Ministral-3-14B-Instruct-2512-GGUF- vLLM
vllm serve mistralai/Ministral-3-14B-Instruct-2512- MLX (Mac)
mlx_lm.chat --model mlx-community/Ministral-3-14B-Instruct-2512-4bit
Questions
How much VRAM does Ministral 3 14B need?
Ministral 3 14B needs 7.9 GB for its weights at 4-bit, 10.0 GB with an 8K context and 13.7 GB with 32K. At 8-bit, 15.9 GB.
Can Ministral 3 14B run on an RTX 4090?
Yes. At 8-bit it holds 48K of context and writes about 47 tokens a second.