Magistral Small 2507
23.6B parameters · 40k context · Apache 2.0 · released 2025-07-18
reasoningchatagentic
In short
Magistral Small 2507 is an open model from Mistral AI with 23.6B parameters. At Q4_K_M, Magistral Small 2507 needs about 15 GB of memory at 8k context and 19 GB at 32k, counting the weights, the KV cache and runtime buffers. The cheapest hardware in Germany that runs Magistral Small 2507 fully in memory at 32k context is the AMD Radeon RX 7900 XTX, 1.555 € at eBay, at about 24 tok/s (estimated).
Updated . Speeds are estimates; how they are worked out.
What to run it on in Germany
Fully in memory at 32k context, priced at retailers today.
Cheapest
AMD Radeon RX 7900 XTX24tok/sest.Q5_K_M
1.555 €
Best value
AMD Radeon RX 7900 XTX24tok/sest.Q5_K_M
1.555 €
Fastest
NVIDIA GeForce RTX 509037tok/sest.Q8_0
6.028 €
Mac
Mac Studio M3 Ultra 96GB18tok/sest.Q8_0
5.999 €
All hardware that runs it
Everything we price, best value first. Two cards split the model by layers.
| Hardware | Quant | Fits | Decode | Price | tok/s per 1,000 |
|---|---|---|---|---|---|
| NVIDIA Tesla P40 | Q5_K_M | fits | 10 tok/s | 369 € | 26.3 |
| Intel Arc Pro B60 | Q5_K_M | fits | 10 tok/s | 1.074 € | 9.6 |
| Mac mini M4 32GB | Q4_K_M | fits | 4 tok/s | 1.400 € | 2.9 |
| AMD Radeon RX 7900 XTX | Q5_K_M | fits | 24 tok/s | 1.555 € | 15.4 |
| NVIDIA GeForce RTX 3090 | Q5_K_M | fits | 26 tok/s | 1.850 € | 14.2 |
| AMD Radeon AI PRO R9700 | Q8_0 | fits | 12 tok/s | 1.867 € | 6.2 |
| NVIDIA GeForce RTX 3090 Ti | Q5_K_M | fits | 28 tok/s | 2.500 € | 11.3 |
| Mac mini M4 Pro 48GB | Q8_0 | fits | 6 tok/s | 2.699 € | 2.1 |
| Mac mini M4 Pro 64GB | Q8_0 | fits | 6 tok/s | 2.799 € | 2.1 |
| GMKtec EVO-X2 Ryzen AI Max+ 395 128GB | Q8_0 | fits | 5 tok/s | 3.700 € | 1.2 |
| NVIDIA GeForce RTX 4090 | Q5_K_M | fits | 28 tok/s | 3.955 € | 7.1 |
| Mac Studio M4 Max 64GB | Q8_0 | fits | 12 tok/s | 4.183 € | 2.8 |
| Mac Studio M4 Max 128GB | Q8_0 | fits | 12 tok/s | 5.800 € | 2.0 |
| Mac Studio M3 Ultra 96GB | Q8_0 | fits | 18 tok/s | 5.999 € | 2.9 |
| NVIDIA GeForce RTX 5090 | Q8_0 | fits | 37 tok/s | 6.028 € | 6.1 |
| NVIDIA DGX Spark 128GB | Q8_0 | fits | 6 tok/s | 8.313 € | 0.7 |
| ASUS Ascent GX10 128GB | Q8_0 | fits | 6 tok/s | 8.623 € | 0.6 |
| Mac Studio M3 Ultra 256GB | Q8_0 | fits | 18 tok/s | 15.555 € | 1.1 |
| Mac Studio M3 Ultra 512GB | Q8_0 | fits | 18 tok/s | 18.995 € | 0.9 |
| NVIDIA RTX PRO 6000 Blackwell | Q8_0 | fits | 37 tok/s | 19.076 € | 1.9 |
| Framework Desktop Ryzen AI Max+ 395 128GB | Q8_0 | fits | 5 tok/s | – | – |
| Framework Desktop Ryzen AI Max+ 395 64GB | Q8_0 | fits | 5 tok/s | – | – |
| NVIDIA GeForce RTX 3060 12GB | Q4_K_M | offload | 5 tok/s | 309 € | 15.5 |
| Intel Arc B580 | Q4_K_M | offload | 5 tok/s | 360 € | 13.1 |
| NVIDIA GeForce RTX 3080 10GB | Q4_K_M | offload | 5 tok/s | 500 € | 9.2 |
| AMD Radeon RX 9060 XT 16GB | Q4_K_M | offload | 6 tok/s | 550 € | 10.7 |
| Intel Arc A770 16GB | Q4_K_M | offload | 7 tok/s | 646 € | 11.3 |
| NVIDIA GeForce RTX 4060 Ti 16GB | Q4_K_M | offload | 6 tok/s | 650 € | 9.4 |
| NVIDIA GeForce RTX 5070 | Q4_K_M | offload | 5 tok/s | 699 € | 7.7 |
| NVIDIA GeForce RTX 4070 Super | Q4_K_M | offload | 5 tok/s | 707 € | 7.4 |
| NVIDIA GeForce RTX 5060 Ti 16GB | Q4_K_M | offload | 8 tok/s | 739 € | 10.1 |
| AMD Radeon RX 9070 XT | Q4_K_M | offload | 8 tok/s | 778 € | 10.2 |
| AMD Radeon RX 7600 XT | Q4_K_M | offload | 6 tok/s | 813 € | 6.9 |
| AMD Radeon RX 7900 XT | Q4_K_M | offload | 21 tok/s | 855 € | 24.8 |
| AMD Radeon RX 7800 XT | Q4_K_M | offload | 8 tok/s | 1.002 € | 7.9 |
| NVIDIA GeForce RTX 4080 Super | Q4_K_M | offload | 9 tok/s | 1.100 € | 8.2 |
| NVIDIA GeForce RTX 5070 Ti | Q4_K_M | offload | 10 tok/s | 1.149 € | 8.4 |
| NVIDIA GeForce RTX 4070 Ti Super | Q4_K_M | offload | 9 tok/s | 1.247 € | 7.1 |
| NVIDIA GeForce RTX 5080 | Q4_K_M | offload | 10 tok/s | 1.500 € | 6.5 |
1 other card or machine we price cannot hold it at this context.
Memory needed
Weights, KV cache and runtime buffers at each context length (KV cache at fp16).
| Quant | File | @8k | @32k |
|---|---|---|---|
| Q4_K_Mgguf | 13.3 GB | 15.3 GB | 19.2 GB |
| Q5_K_Mgguf | 15.6 GB | 17.6 GB | 21.5 GB |
| Q8_0gguf | 23.3 GB | 25.3 GB | 29.2 GB |
Run it
Copy a command for the runtime you use.
- Q4_K_M
- Ollama
ollama run hf.co/mistralai/Magistral-Small-2507-GGUF:Q4_K_M - LM Studio
lms get mistralai/Magistral-Small-2507-GGUF@q4_k_m - llama.cpp
llama-server -hf mistralai/Magistral-Small-2507-GGUF:Q4_K_M
- Ollama
- Q5_K_M
- Ollama
ollama run hf.co/mistralai/Magistral-Small-2507-GGUF:Q5_K_M - LM Studio
lms get mistralai/Magistral-Small-2507-GGUF@q5_k_m - llama.cpp
llama-server -hf mistralai/Magistral-Small-2507-GGUF:Q5_K_M
- Ollama
- Q8_0
- Ollama
ollama run hf.co/mistralai/Magistral-Small-2507-GGUF:Q8_0 - LM Studio
lms get mistralai/Magistral-Small-2507-GGUF@q8_0 - llama.cpp
llama-server -hf mistralai/Magistral-Small-2507-GGUF:Q8_0
- Ollama
Notes
128k context window, but performance may degrade past 40k, so a 40k max model length is recommended. Needs mistral_common >= 1.8.2 and the latest vLLM.
Questions
- How much VRAM does Magistral Small 2507 need?
- At Q4_K_M, Magistral Small 2507 needs about 15 GB of memory at 8k context and 19 GB at 32k, counting the weights, the KV cache and runtime buffers.
- What is the best GPU for running Magistral Small 2507 locally?
- The cheapest hardware in Germany that runs Magistral Small 2507 fully in memory at 32k context is the AMD Radeon RX 7900 XTX, 1.555 € at eBay, at about 24 tok/s (estimated). The fastest we price is the NVIDIA GeForce RTX 5090, at about 37 tok/s.
- Can I run Magistral Small 2507 on a 16 GB GPU?
- Only partly: at 32k context it does not fit in 16 GB, so part of it runs from system RAM and it slows to about 6 tok/s on the AMD Radeon RX 9060 XT 16GB (estimated).
- Can I run Magistral Small 2507 on a 24 GB GPU?
- Yes: at Q5_K_M it fits fully in 24 GB at 32k context. On the NVIDIA GeForce RTX 3090 Ti it runs at about 28 tok/s (estimated).
- How do I run Magistral Small 2507 locally?
- With Ollama: ollama run hf.co/mistralai/Magistral-Small-2507-GGUF:Q4_K_M (the Q4_K_M build). LM Studio and llama.cpp commands are on this page.
- What licence is Magistral Small 2507 released under?
- Apache 2.0, which allows commercial use.