readygpu

Ministral 3 8B Reasoning 2512

8.9B parameters · 256k context · Apache 2.0 · released 2025-10-31

chatcodingagenticvisionreasoning

In short

Ministral 3 8B Reasoning 2512 is an open model from Mistral AI with 8.9B parameters. At Q4_K_M, Ministral 3 8B Reasoning 2512 needs about 7 GB of memory at 8k context and 10 GB at 32k, counting the weights, the KV cache and runtime buffers. The cheapest hardware in the US that runs Ministral 3 8B Reasoning 2512 fully in memory at 32k context is the NVIDIA Tesla P40, $345 at Amazon, at about 16 tok/s (estimated).

Updated . Speeds are estimates; how they are worked out.

What to run it on in the US

Fully in memory at 32k context, priced at retailers today.

All hardware that runs it

Everything we price, best value first. Two cards split the model by layers.

HardwareQuantFitsDecodePricetok/s per 1,000
NVIDIA Tesla P40Q8_0fits16 tok/s$34545.8
Intel Arc B580Q5_K_Mfits22 tok/s$34662.1
NVIDIA GeForce RTX 3060 12GBQ5_K_Mfits21 tok/s$36058.3
Intel Arc A770 16GBQ8_0fits21 tok/s$40051.5
AMD Radeon RX 7600 XTQ8_0fits12 tok/s$48024.4
AMD Radeon RX 9060 XT 16GBQ8_0fits13 tok/s$52024.8
AMD Radeon RX 7800 XTQ8_0fits25 tok/s$55045.8
NVIDIA GeForce RTX 5060 Ti 16GBQ8_0fits20 tok/s$61833.0
Intel Arc Pro B60Q8_0fits17 tok/s$65025.8
NVIDIA GeForce RTX 4070 SuperQ5_K_Mfits29 tok/s$69942.0
Mac mini M4 16GBQ5_K_Mfits7 tok/s$7499.7
NVIDIA GeForce RTX 5070Q5_K_Mfits39 tok/s$75052.3
NVIDIA GeForce RTX 4060 Ti 16GBQ8_0fits13 tok/s$77017.0
AMD Radeon RX 7900 XTQ8_0fits32 tok/s$80040.5
AMD Radeon RX 9070 XTQ8_0fits26 tok/s$81531.8
NVIDIA GeForce RTX 4070 Ti SuperQ8_0fits31 tok/s$98431.2
NVIDIA GeForce RTX 5070 TiQ8_0fits41 tok/s$99541.1
NVIDIA GeForce RTX 4080 SuperQ8_0fits34 tok/s$1,19928.0
Mac mini M4 32GBQ8_0fits6 tok/s$1,4204.0
AMD Radeon RX 7900 XTXQ8_0fits39 tok/s$1,53325.3
NVIDIA GeForce RTX 3090Q8_0fits43 tok/s$1,60026.7
NVIDIA GeForce RTX 5080Q8_0fits44 tok/s$1,60027.4
AMD Radeon AI PRO R9700Q8_0fits26 tok/s$1,70015.2
NVIDIA GeForce RTX 3090 TiQ8_0fits46 tok/s$2,06022.3
NVIDIA GeForce RTX 4090Q8_0fits46 tok/s$2,07322.2
Mac mini M4 Pro 48GBQ8_0fits13 tok/s$2,1496.1
Mac mini M4 Pro 64GBQ8_0fits13 tok/s$2,9004.5
Mac Studio M4 Max 64GBQ8_0fits26 tok/s$3,4007.7
Framework Desktop Ryzen AI Max+ 395 128GBQ8_0fits10 tok/s$3,4493.0
GMKtec EVO-X2 Ryzen AI Max+ 395 128GBQ8_0fits10 tok/s$3,6502.8
Mac Studio M3 Ultra 96GBQ8_0fits39 tok/s$4,3998.9
NVIDIA GeForce RTX 5090Q8_0fits82 tok/s$4,92116.6
Mac Studio M4 Max 128GBQ8_0fits26 tok/s$4,9995.2
ASUS Ascent GX10 128GBQ8_0fits13 tok/s$6,1002.0
NVIDIA DGX Spark 128GBQ8_0fits13 tok/s$7,3991.7
NVIDIA RTX PRO 6000 BlackwellQ8_0fits82 tok/s$7,99710.2
Mac Studio M3 Ultra 256GBQ8_0fits39 tok/s$8,4994.6
Mac Studio M3 Ultra 512GBQ8_0fits39 tok/s$23,0001.7
Framework Desktop Ryzen AI Max+ 395 64GBQ8_0fits10 tok/s––
NVIDIA GeForce RTX 3080 10GBQ4_K_Moffload32 tok/s$46968.8

Memory needed

Weights, KV cache and runtime buffers at each context length (KV cache at fp16).

QuantFile@8k@32k@64k@128k
Q4_K_Mgguf4.8 GB6.7 GB10.0 GB14.4 GB23.3 GB
Q5_K_Mgguf5.6 GB7.5 GB10.8 GB15.2 GB24.1 GB
Q8_0gguf8.4 GB10.2 GB13.5 GB18.0 GB26.9 GB

Run it

Copy a command for the runtime you use.

  • Q4_K_M
    • Ollamaollama run hf.co/mistralai/Ministral-3-8B-Reasoning-2512-GGUF:Q4_K_M
    • LM Studiolms get mistralai/Ministral-3-8B-Reasoning-2512-GGUF@q4_k_m
    • llama.cppllama-server -hf mistralai/Ministral-3-8B-Reasoning-2512-GGUF:Q4_K_M
  • Q5_K_M
    • Ollamaollama run hf.co/mistralai/Ministral-3-8B-Reasoning-2512-GGUF:Q5_K_M
    • LM Studiolms get mistralai/Ministral-3-8B-Reasoning-2512-GGUF@q5_k_m
    • llama.cppllama-server -hf mistralai/Ministral-3-8B-Reasoning-2512-GGUF:Q5_K_M
  • Q8_0
    • Ollamaollama run hf.co/ggml-org/Ministral-3-8B-Reasoning-2512-GGUF:Q8_0
    • LM Studiolms get ggml-org/Ministral-3-8B-Reasoning-2512-GGUF@q8_0
    • llama.cppllama-server -hf ggml-org/Ministral-3-8B-Reasoning-2512-GGUF:Q8_0

Notes

Needs vllm >= 0.12.0 (with mistral_common >= 1.8.6) or Transformers 5.0.0rc0. Fits in 24GB of VRAM in BF16, or under 12GB when quantized.

Questions

How much VRAM does Ministral 3 8B Reasoning 2512 need?
At Q4_K_M, Ministral 3 8B Reasoning 2512 needs about 7 GB of memory at 8k context and 10 GB at 32k, counting the weights, the KV cache and runtime buffers.
What is the best GPU for running Ministral 3 8B Reasoning 2512 locally?
The cheapest hardware in the US that runs Ministral 3 8B Reasoning 2512 fully in memory at 32k context is the NVIDIA Tesla P40, $345 at Amazon, at about 16 tok/s (estimated). The best value is the Intel Arc B580 (about 22 tok/s for $346). The fastest we price is the NVIDIA GeForce RTX 5090, at about 82 tok/s.
Can I run Ministral 3 8B Reasoning 2512 on a 16 GB GPU?
Yes: at Q8_0 it fits fully in 16 GB at 32k context. On the NVIDIA GeForce RTX 5080 it runs at about 44 tok/s (estimated).
Can I run Ministral 3 8B Reasoning 2512 on a 24 GB GPU?
Yes: at Q8_0 it fits fully in 24 GB at 32k context. On the NVIDIA GeForce RTX 3090 Ti it runs at about 46 tok/s (estimated).
How do I run Ministral 3 8B Reasoning 2512 locally?
With Ollama: ollama run hf.co/mistralai/Ministral-3-8B-Reasoning-2512-GGUF:Q4_K_M (the Q4_K_M build). LM Studio and llama.cpp commands are on this page.
What licence is Ministral 3 8B Reasoning 2512 released under?
Apache 2.0, which allows commercial use.