1344 GB/s (+544)
C
Raises estimated decode speed by about 143%.77.1 tok/s decode
Raises estimated decode speed by about 143%.
~$4,999 MSRP
Can Mistral Small 3.2 24B Instruct 2506 run on Mac Studio M2 Ultra 64GB?
YES — Runs Great
C50Usable
○Estimated from fit model
Mistral Small 3.2 24B Instruct 2506 needs ~25.3 GB VRAM. Mac Studio M2 Ultra 64GB has 46.1 GB. With Q4_K_M quantization, expect ~32 tok/s.
Runtime: llama.cppCapacity: RoomyBandwidth: HighStack: StandardBottleneck: Balanced
Share:
Operating mode
Choose the run profile you care about
Interactive favors responsiveness, while light API and scale-out lean harder on serving readiness. The fit stays the same, but the recommendation lens changes.
Current mode
Balanced
Balanced for general local use. Keeps the ranking neutral across personal and serving workflows.
Select quantization to explore
Q4_K_M (Medium quality) — 25.3 GB, 31.7 tok/s, Runs well
25.3 GB required46.1 GB available
55% VRAM used
Fit status
Runs well
Decode
31.7 tok/s
TTFT
6108 ms
Safe context
134K
Memory
25.3 GB / 46.1 GB
Memory breakdown
Weights14.6 GB
KV Cache2.8 GB
Runtime0.9 GB
Headroom6.9 GB
See how fast it feels
See how fast it feelsMistral Small 3.2 24B Instruct 2506 on Mac Studio M2 Ultra 64GB
1st promptCold start — includes initialization
>What is local AI inference?
Local AI inference means running an AI language model directly on your own hardware — your laptop, desktop, or server — instead of sending requests to a remote cloud API.
When you run inference locally the model weights are loaded into your GPU or unified memory. Each token you generate requires reading those weights from memory, so memory bandwidth is the main bottleneck for decode speed.
Key benefits of running locally:
- Full privacy: your prompts never leave your machine
- No per-token cost or rate limits
- Works offline once the model is downloaded
- Latency depends only on your hardware
2nd promptWarm — model ready, no init overhead
>How much VRAM do I need?
It depends on the model size and quantization level. A rough rule of thumb:
Model size Q4 (4-bit) Q8 (8-bit) FP16
7B params ~4.3 GB ~7.5 GB ~14 GB
13B params ~7.9 GB ~13.9 GB ~26 GB
70B params ~42.7 GB ~74.9 GB ~140 GB
Most people use 4-bit quantization (Q4_K_M) which gives 90-95% of full quality at a fraction of the memory. A 24 GB GPU can comfortably run most 7B-13B models.
Estimated: 31.7 tok/s decode · 6.1s TTFT (warm) · 79 tok/s prefill
What limits this setup
This setup is broadly balanced for this model.
Shared-memory contention still exists
The OS, browser, and inference runtime all compete for the same physical memory pool, so real-world headroom is less forgiving than raw capacity suggests.
Best improvement path
Performance by workload
| Workload | Grade | Fit | Decode | TTFT | Context |
|---|---|---|---|---|---|
| Chat | C | Runs well | 31.7 tok/s | 3332 ms | 134K |
| Coding | C | Runs well | 31.7 tok/s | 6108 ms | 134K |
| Agentic Coding | C | Runs well | 31.7 tok/s | 8885 ms | 134K |
| Reasoning | C | Runs well | 31.7 tok/s | 7219 ms | 134K |
| RAG | C | Runs well | 31.7 tok/s | 11106 ms | 134K |
Quantization options
How Mistral Small 3.2 24B Instruct 2506 (24B params) fits at each quantization level on Mac Studio M2 Ultra 64GB (46.1 GB usable).
| Quant | Bits | VRAM | Quality | Fit |
|---|---|---|---|---|
Q2_K | 2 | 9.4 GB | Low | C43 |
Q3_K_S | 3 | 11.8 GB | Low | C44 |
NVFP4 | 4 |
Get started
Copy-paste commands to run Mistral Small 3.2 24B Instruct 2506 on your machine.
Run
lms load hf-unsloth--mistral-small-3-2-24b-instruct-2506-gguf && lms server startUpgrade options