unsloth
Ministral 3 3B Reasoning 2512
Ministral 3 3B Reasoning 2512 is a 3.43B dense mistral3 model from unsloth. Configured context length is 262,144 tokens. License: apache-2.0. Quant sizes are the GGUF file bytes published on Hugging Face when a file exists.
Parameters3.4B
Context256K
Attentiongqa
Licenseapache-2.0
26 layers · 8 KV heads · head dim 128 · GGUF repo · config
Quant file sizes
| Quant | File size | Source |
|---|---|---|
| BF16 | 6.39 GB | Hugging Face file bytes |
| UD-Q8_K_XL | 4.19 GB | Hugging Face file bytes |
| Q8_0 | 3.40 GB | Hugging Face file bytes |
| UD-Q6_K_XL | 2.96 GB | Hugging Face file bytes |
| Q6_K | 2.63 GB | Hugging Face file bytes |
| UD-Q5_K_XL | 2.31 GB | Hugging Face file bytes |
| Q5_K_M | 2.30 GB | Hugging Face file bytes |
| Q5_K_S | 2.25 GB | Hugging Face file bytes |
| UD-Q4_K_XL | 2.04 GB | Hugging Face file bytes |
| Q4_K_M | 2.00 GB | Hugging Face file bytes |
| Q4_K_S | 1.91 GB | Hugging Face file bytes |
| IQ4_NL | 1.91 GB | Hugging Face file bytes |
| IQ4_XS | 1.82 GB | Hugging Face file bytes |
| Q4_1 | 2.08 GB | Hugging Face file bytes |
| Q4_0 | 1.91 GB | Hugging Face file bytes |
| UD-Q3_K_XL | 1.73 GB | Hugging Face file bytes |
| Q3_K_M | 1.67 GB | Hugging Face file bytes |
| Q3_K_S | 1.53 GB | Hugging Face file bytes |
| UD-IQ3_XXS | 1.36 GB | Hugging Face file bytes |
| UD-Q2_K_XL | 1.39 GB | Hugging Face file bytes |
| Q2_K | 2.72 GB | Hugging Face file bytes |
| UD-IQ2_XXS | 1.03 GB | Hugging Face file bytes |
| UD-IQ1_M | 0.95 GB | Hugging Face file bytes |
| UD-IQ1_S | 0.90 GB | Hugging Face file bytes |
On common machines
| Device | Verdict | Quant | Speed |
|---|---|---|---|
| GeForce RTX 4060 Ti 8GB | Runs | UD-Q8_K_XL | 29–40 |
| GeForce RTX 3060 12GB | Runs | BF16 | 26–35 |
| GeForce RTX 4060 Ti 16GB | Runs great | BF16 | 20–28 |
| GeForce RTX 4090 | Runs great | BF16 | 72–98 |
| GeForce RTX 5090 | Runs great | BF16 | 127–174 |
| Apple M4 Pro | Runs great | BF16 | 19–26 |
| Apple M5 | Runs | BF16 | 11–15 |
| CPU only · DDR5-5600 | Runs on CPU | BF16 | 6.4–8.7 |
Smallest devices that fit it fully
- GeForce RTX 3050 6GB — Tight fit, UD-Q8_K_XL, 19–26 tok/s
- GeForce RTX 2060 6GB — Tight fit, UD-Q8_K_XL, 37–51 tok/s
- GeForce GTX 1660 Super — Tight fit, UD-Q8_K_XL, 37–51 tok/s
- GeForce GTX 1660 Ti — Tight fit, UD-Q8_K_XL, 32–44 tok/s
- Intel Arc A380 — Tight fit, UD-Q8_K_XL, 21–28 tok/s
- GeForce RTX 5060 Ti 8GB — Runs, UD-Q8_K_XL, 46–63 tok/s
Run it
Commands for the GeForce RTX 4060 Ti 16GB estimate (BF16).
llama.cpp
llama-server -hf unsloth/Ministral-3-3B-Reasoning-2512-GGUF:BF16 -c 8192 -ngl 26
No matching name in the Ollama library snapshot.
LM Studio
lms get unsloth/Ministral-3-3B-Reasoning-2512-GGUF # Then load quant BF16