Mistral AI · Apache 2.0
Ministral 3
Mistral’s small models for laptops and 12–16 GB cards, with image input and a 256K context.
How much memory it needs
The model file, plus memory for the conversation, plus about 1 GB for the app. Longer conversations need more.
| Size | Version | File | At 8K | At 32K | At 128K |
|---|---|---|---|---|---|
| 3.8B | Q4_K_M | 2.1 GB | 4.9 GB | 7.5 GB | 17.9 GB |
| Q8_0 | 3.6 GB | 6.4 GB | 9.0 GB | 19.4 GB | |
| 8.9B | Q4_K_M | 5.2 GB | 8.2 GB | 11.6 GB | 25.3 GB |
| Q8_0 | 9.0 GB | 12.0 GB | 15.4 GB | 29.1 GB | |
| 14B | Q4_K_M | 8.2 GB | 11.5 GB | 15.5 GB | 31.6 GB |
| Q8_0 | 14.4 GB | 17.6 GB | 21.6 GB | 37.7 GB |
Which computers run it
The best size for each machine at an 8K-token conversation. Speeds are estimates from memory bandwidth.
| Computer | Best version | Fit | Speed |
|---|---|---|---|
| Laptop or PC, no graphics card, 32 GB | 14B Q4_K_M | Runs great | 5.5–9tok/sReadable |
| RTX 4060 · 8 GB | 3.8B Q8_0 | Runs well | 35–55tok/sFast |
| RTX 3060 · 12 GB | 8.9B Q4_K_M | Runs great | 35–55tok/sFast |
| RTX 5060 Ti · 16 GB | 14B Q4_K_M | Runs great | 30–45tok/sSmooth |
| RTX 4090 · 24 GB | 14B Q8_0 | Runs great | 35–60tok/sFast |
| RTX 5090 · 32 GB | 14B Q8_0 | Runs great | 65–100tok/sFast |
| Mac with M4 · 16 GB | 8.9B Q4_K_M | Runs great | 11–19tok/sSmooth |
| Mac with M4 Pro · 48 GB | 14B Q8_0 | Runs great | 10–17tok/sSmooth |
| Mac with M5 Max · 128 GB | 14B Q8_0 | Runs great | 20–40tok/sSmooth |
| Ryzen AI Max+ 395 · 128 GB | 14B Q8_0 | Runs great | 10–15tok/sReadable |
| DGX Spark · 128 GB | 14B Q8_0 | Runs great | 10–17tok/sReadable |
| RTX PRO 6000 · 96 GB | 14B Q8_0 | Runs great | 65–100tok/sFast |
| Mac Studio with M5 Ultra · 256 GB | 14B Q8_0 | Runs great | 35–55tok/sFast |
| Mac Studio with M3 Ultra · 512 GB | 14B Q8_0 | Runs great | 25–40tok/sSmooth |
What it is good and bad at
Good at
- Fits small machines
- Images and tool calls
- Long 256K context
Not good at
- Behind Qwen3.5 of the same size on hard tasks