How much VRAM does Mistral Small 3.2 24B need?
Mistral Small 3.2 24B needs 13.6 GB for its weights at 4-bit, 15.7 GB with an 8K context and 19.4 GB with 32K. At 8-bit, 25.9 GB.
- Parameters
- 24B
- Longest context
- 128K
- Released
- Jun 2025
- License
- Apache 2.0
- Strongest at
- Chat, Coding, Agents
Memory at each precision
| Precision | Weights | With 8K | With 32K | With 128K |
|---|---|---|---|---|
| 16-bit | 44.7 GB | 46.8 GB | 50.6 GB | 65.6 GB |
| 8-bit | 23.8 GB | 25.9 GB | 29.6 GB | 44.6 GB |
| 6-bit | 18.3 GB | 20.4 GB | 24.2 GB | 39.2 GB |
| 5-bit | 15.9 GB | 18.0 GB | 21.8 GB | 36.8 GB |
| 4-bit | 13.6 GB | 15.7 GB | 19.4 GB | 34.4 GB |
| 3-bit | 10.9 GB | 13.0 GB | 16.8 GB | 31.8 GB |
| 2-bit | 8.4 GB | 10.5 GB | 14.2 GB | 29.2 GB |
| 1.6-bit | 4.6 GB | 6.7 GB | 10.5 GB | 25.5 GB |
One conversation, 16-bit context cache, runtime overhead included.
What runs it
| Machine | Best fit | Context | Speed | First reply |
|---|---|---|---|---|
| RTX 3060 12 GB | Too big | |||
| RTX 4060 Ti 16 GB | 3-bit | 20K | 17 tok/s | 2.3 s |
| RTX 3090 | 6-bit | 20K | 33 tok/s | 1.4 s |
| RTX 4090 | 6-bit | 20K | 35 tok/s | 0.6 s |
| RTX 5090 | 8-bit | 36K | 48 tok/s | 0.5 s |
| 2× RTX 3090 | 8-bit | 128K | 26 tok/s | 0.8 s |
| RTX PRO 6000 Blackwell | 8-bit | 128K | 48 tok/s | 0.4 s |
| M4 Max 128 GB | 8-bit | 128K | 14 tok/s | 7.9 s |
| M5 Max 128 GB | 8-bit | 128K | 15 tok/s | 2.3 s |
| Ryzen AI Max+ 395 128 GB | 8-bit | 128K | 6.1 tok/s | 4.5 s |
| DGX Spark 128 GB | 8-bit | 128K | 7.6 tok/s | 1.6 s |
| M3 Ultra 256 GB | 8-bit | 128K | 20 tok/s | 4.4 s |
| M5 Ultra 512 GB | 8-bit | 128K | 29 tok/s | 1.4 s |
| H100 SXM | 8-bit | 128K | 85 tok/s | 0.3 s |
| DGX Station GB300 | 8-bit | 128K | 162 tok/s | 0.1 s |
| 8× H200 NVL | 8-bit | 128K | 117 tok/s | 0.1 s |
Best fit is the highest quality, up to 8-bit, that leaves 16K of context. First reply is the wait for a 2K prompt. Speeds are rough, plus or minus 25%.
How to run it
- Ollama
ollama run mistral-small3.2:24b- llama.cpp
llama-server -hf unsloth/Mistral-Small-3.2-24B-Instruct-2506-GGUF- vLLM
vllm serve mistralai/Mistral-Small-3.2-24B-Instruct-2506- MLX (Mac)
mlx_lm.chat --model mlx-community/Mistral-Small-3.2-24B-Instruct-2506-4bit
Questions
How much VRAM does Mistral Small 3.2 24B need?
Mistral Small 3.2 24B needs 13.6 GB for its weights at 4-bit, 15.7 GB with an 8K context and 19.4 GB with 32K. At 8-bit, 25.9 GB.
Can Mistral Small 3.2 24B run on an RTX 4090?
Yes. At 6-bit it holds 20K of context and writes about 35 tokens a second.