Mistral AI · llama
Can I run magistral?
Magistral is a small, efficient reasoning model with 24B parameters.
23.6B parameters39K contexttextApache License3 builds
magistral system requirements
Memory needed at 4K context, counting model weights, KV cache, compute buffers and runtime overhead. Download sizes come from the official Ollama registry.
| Build | Quantization | Download | Memory needed |
|---|---|---|---|
| magistral:24b | Q4_K_M | 13.3 GB | ~19.5 GB(est.) |
| magistral:24b-small-2506-q8_0 | Q8_0 | 23.3 GB | ~33.8 GB(est.) |
| magistral:24b-small-2506-fp16 | F16 | 43.9 GB | ~63.2 GB(est.) |
Figures marked (est.) approximate the KV cache because this model's architecture details are not published. They are indicative rather than exact.
How to run magistral locally
- 1.Install Ollama for Windows, macOS or Linux.
- 2.Run this in a terminal:
ollama run magistral:24b - 3.The weights download on first run, then an interactive prompt opens.
Frequently asked questions about magistral
- How much VRAM do I need to run magistral?
- At 4K context, the smallest published build of magistral (Q4_K_M) needs roughly 19.5 GB of GPU memory once weights, KV cache, compute buffers and runtime overhead are counted. Bigger quantizations and longer context windows need more. It can also run partly or entirely on the CPU using system RAM, more slowly.
- Can I run magistral without a dedicated GPU?
- Yes, but slowly. Without a GPU the model runs on the CPU using system RAM, which usually means a few tokens per second rather than dozens. You would need at least 19.5 GB of free RAM for the smallest build.
- How big is the magistral download?
- The smallest published build is 13.3 GB. There are 3 builds in total, the largest being 43.9 GB. Leave some extra free disk space beyond the download itself.
- How do I run magistral locally?
- Install Ollama, then run "ollama run magistral:24b" in a terminal. The weights download on first use and an interactive session opens.
Will magistral run on your PC?
Scan your hardware once and get a verdict for this model and every build — free, no account.
Model data from the official Ollama registry · verified 13/09/2026