# Mistral Small 3.2 24B Instruct (2506)

> OnPrem World · https://onpremworld.com/en/models/mistral-small-3.2-24b · Last updated: 2026-10-14 · CC BY 4.0

Mistral's 24B with image input and function calling, under Apache-2.0; it fits a 16 GB card at 4-bit.

|  |  |
|---|---|
| Developer | Mistral AI |
| Parameters | 24B |
| Native context | 131,072 tokens |
| Licence | Apache-2.0 |
| Commercial use | Yes |
| KV cache | All 40 layers (standard attention) |
| Released | 2025-06 |

```
ollama run mistral-small3.2:24b
```

## Memory needed

Weights + KV cache + runtime overhead, in GiB. Lower quantization = less memory, slightly lower quality.

| Quantization | 4K | 8K | 32K |
|---|---|---|---|
| Q4_K_M | 15.4 GiB | 16.0 GiB | 19.7 GiB |
| Q5_K_M | 17.8 GiB | 18.4 GiB | 22.2 GiB |
| Q6_K | 20.5 GiB | 21.1 GiB | 24.8 GiB |
| Q8_0 | 26.1 GiB | 26.7 GiB | 30.4 GiB |
| FP16 | 48.1 GiB | 48.7 GiB | 52.4 GiB |

## Which hardware runs it (8K)

| Hardware | Memory | Best quant | tok/s (est.) |
|---|---|---|---|
| GeForce RTX 5090 | 32 GB | Q8_0 | 49 |
| GeForce RTX 5080 | 16 GB | Q4_K_M | 46 |
| GeForce RTX 5070 Ti | 16 GB | Q4_K_M | 43 |
| GeForce RTX 4090 | 24 GB | Q6_K | 36 |
| GeForce RTX 4080 SUPER | 16 GB | Q4_K_M | 35 |
| Radeon RX 7900 XTX | 24 GB | Q6_K | 34 |
| GeForce RTX 3090 | 24 GB | Q6_K | 33 |
| Radeon RX 7900 XT | 20 GB | Q5_K_M | 33 |
| Radeon RX 9070 XT | 16 GB | Q4_K_M | 31 |
| RTX 6000 Ada Generation | 48 GB | Q8_0 | 26 |
| RTX PRO 6000 Blackwell Workstation Edition | 96 GB | FP16 | 26 |
| GeForce RTX 5060 Ti 16GB | 16 GB | Q4_K_M | 22 |
| RTX A6000 | 48 GB | Q8_0 | 21 |
| RTX PRO 5000 72GB Blackwell | 72 GB | FP16 | 20 |
| Radeon AI PRO R9700 | 32 GB | Q8_0 | 18 |
| Arc Pro B70 | 32 GB | Q8_0 | 17 |
| M5 Ultra | 512 GB | FP16 | 16 |
| Arc Pro B60 | 24 GB | Q6_K | 16 |
| Radeon RX 9060 XT 16GB | 16 GB | Q4_K_M | 15 |
| GeForce RTX 4060 Ti 16GB | 16 GB | Q4_K_M | 14 |
| M5 Max (18-core CPU / 32-core GPU) | 36 GB | Q8_0 | 12 |
| M3 Ultra | 512 GB | FP16 | 11 |
| M4 Max (14-core CPU / 32-core GPU) | 36 GB | Q8_0 | 10 |
| M5 Max (18-core CPU / 40-core GPU) | 128 GB | FP16 | 8.3 |
| M5 Pro | 64 GB | Q8_0 | 7.8 |
| M4 Max (16-core CPU / 40-core GPU) | 128 GB | FP16 | 7.4 |
| M4 Pro | 64 GB | Q8_0 | 7.0 |
| M6 (24GB/32GB configurations, 170 GB/s) | 32 GB | Q6_K | 5.6 |
| M5 | 32 GB | Q6_K | 5.0 |
| RTX Spark (N1X) laptops & compact desktops | 128 GB | FP16 | 3.8 |
| DGX Spark | 128 GB | FP16 | 3.4 |
| Ryzen AI Max+ PRO 495 (Gorgon Halo) systems | 192 GB | FP16 | 3.4 |
| Ryzen AI Max+ 395 (Strix Halo) systems | 128 GB | FP16 | 3.2 |
| GeForce RTX 3060 (12 GB) | 12 GB | Doesn't fit | – |
| GeForce RTX 5070 | 12 GB | Doesn't fit | – |
| Arc B580 | 12 GB | Doesn't fit | – |
| M6 (16GB configuration, 153 GB/s) | 16 GB | Doesn't fit | – |

## Frequently asked questions

### How much VRAM does Mistral Small 3.2 24B Instruct (2506) need?

With an 8K context, about 16.0 GiB at Q4_K_M and 26.7 GiB at Q8_0 (weights + KV cache + runtime overhead). Longer contexts need more memory for the KV cache.

### Can Mistral Small 3.2 24B Instruct (2506) run on an GeForce RTX 4090?

Yes. Mistral Small 3.2 24B Instruct (2506) fits on the GeForce RTX 4090 (24 GB) at Q6_K with an 8K context, using about 21.1 GiB, at an estimated 36 tokens/s.

### Can I run Mistral Small 3.2 24B Instruct (2506) on a Mac?

Yes. Mistral Small 3.2 24B Instruct (2506) fits on the M3 Ultra (96 GB) at FP16 with an 8K context, using about 48.7 GiB, at an estimated 11 tokens/s.

### Can Mistral Small 3.2 24B Instruct (2506) be used commercially?

Mistral Small 3.2 24B Instruct (2506) is released under Apache-2.0. The licence allows commercial use (read its conditions).

## Sources

- https://docs.mistral.ai/getting-started/changelog/
- https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506
- https://huggingface.co/mistralai/Mistral-Small-3.2-24B-Instruct-2506/resolve/main/config.json
- https://huggingface.co/api/models/mistralai/Mistral-Small-3.2-24B-Instruct-2506?expand[]=safetensors
- https://ollama.com/library/mistral-small3.2/tags