# M3 Ultra

> OnPrem World · https://onpremworld.com/en/hardware/apple-m3-ultra · Last updated: 2026-10-14 · CC BY 4.0

819 GB/s and 96–256 GB of unified memory in the Mac Studio (512 GB at launch, since dropped); 256 GB holds 280B-class MoE models like DeepSeek-V4-Flash at 4-bit.

## Specifications

|  |  |
|---|---|
| Memory | 96–512 GB LPDDR5 |
| Memory bandwidth | 819 GB/s |
| Launch | 2025-03 |
| Launch price | US$3,999 |
| Architecture | Apple M3 family (3 nm; TSMC N3B), two M3 Max dies joined by UltraFusion |
| Software | Metal / MLX |

## What it can run (512 GB, 8K)

Estimates, not benchmarks. See the methodology.

| Model | Parameters | Best quant | Total | tok/s (est.) |
|---|---|---|---|---|
| DeepSeek-V4-Flash | 284B | Q8_0 | 295.6 GiB | 39 |
| Qwen3.8-Flash-Next | 180B | FP16 | 352.7 GiB | 44 |
| Qwen3.5-122B-A10B | 122B | FP16 | 239.3 GiB | 27 |
| Mistral Small 4 119B (2603) | 119B | FP16 | 233.4 GiB | 41 |
| gpt-oss-120b | 117B | FP16 | 229.6 GiB | 52 |
| Llama 4 Scout 17B-16E Instruct | 109B | FP16 | 215.2 GiB | 16 |
| GLM-4.5-Air | 106B | FP16 | 209.2 GiB | 22 |
| Qwen3-Coder-Next (80B-A3B) | 80B | FP16 | 157.1 GiB | 89 |
| Llama 3.3 70B Instruct | 70B | FP16 | 139.9 GiB | 3.8 |
| Qwen3.5-35B-A3B | 35B | FP16 | 69.1 GiB | 89 |
| Qwen3.6-35B-A3B | 35B | FP16 | 69.1 GiB | 89 |
| Qwen3-32B | 32.8B | FP16 | 66.6 GiB | 8.1 |
| DeepSeek-R1-Distill-Qwen-32B | 32.76B | FP16 | 66.6 GiB | 8.1 |
| Gemma 4 31B IT | 30.7B | FP16 | 61.6 GiB | 8.7 |
| Qwen3-30B-A3B (2507) | 30.5B | FP16 | 60.9 GiB | 81 |
| Qwen3-Coder-30B-A3B-Instruct | 30.5B | FP16 | 60.9 GiB | 81 |
| GLM-4.7-Flash (30B-A3B) | 30B | FP16 | 59.6 GiB | 89 |
| Qwen3.5-27B | 27B | FP16 | 53.8 GiB | 9.9 |
| Qwen3.6-27B | 27B | FP16 | 53.8 GiB | 9.9 |
| Qwen3.8-27B | 27B | FP16 | 53.8 GiB | 9.9 |
| Gemma 3 27B IT | 27B | FP16 | 54.3 GiB | 9.9 |
| Gemma 4 26B A4B IT | 25.2B | FP16 | 50.1 GiB | 70 |
| Mistral Small 3.2 24B Instruct (2506) | 24B | FP16 | 48.7 GiB | 11 |
| Magistral Small 1.2 (2509) | 24B | FP16 | 48.7 GiB | 11 |
| Devstral Small 2 24B Instruct (2512) | 24B | FP16 | 48.7 GiB | 11 |
| gpt-oss-20b | 21B | FP16 | 41.8 GiB | 74 |
| Phi-4 (14B) | 14B | FP16 | 29.4 GiB | 19 |
| Gemma 3 12B IT | 12B | FP16 | 24.8 GiB | 22 |
| Gemma 4 12B Unified IT | 11.95B | FP16 | 24.2 GiB | 22 |
| Qwen3.5-9B | 9B | FP16 | 18.4 GiB | 30 |
| Qwen3-8B | 8.2B | FP16 | 17.7 GiB | 32 |
| DeepSeek-R1-0528-Qwen3-8B | 8.19B | FP16 | 17.6 GiB | 33 |
| Llama 3.1 8B Instruct | 8B | FP16 | 17.1 GiB | 33 |
| Gemma 4 E4B IT | 8B | FP16 | 16.4 GiB | 33 |
| Llama 3.2 3B Instruct | 3.21B | FP16 | 7.7 GiB | 83 |
| Kimi K2.6 | 1000B | Doesn't fit | – | – |
| DeepSeek-R1-0528 (671B) | 671B | Doesn't fit | – | – |

## Prices

No prices recorded for this device yet.

## Frequently asked questions

### Which LLMs can the M3 Ultra run?

With 512 GB and an 8K context, the largest models that fit include: DeepSeek-V4-Flash, Qwen3.8-Flash-Next, Qwen3.5-122B-A10B, Mistral Small 4 119B (2603), gpt-oss-120b. See the full table below.

### How fast is the M3 Ultra for local LLMs?

Generation speed is mostly set by memory bandwidth (819 GB/s). For Llama 3.1 8B Instruct at FP16, expect roughly 33 tokens/s; bigger models are slower.

### Can the M3 Ultra run 70B models?

Yes. Llama 3.3 70B Instruct fits on the M3 Ultra (512 GB) at FP16 with an 8K context, using about 139.9 GiB, at an estimated 3.8 tokens/s.

## Sources

- https://www.apple.com/newsroom/2025/03/apple-reveals-m3-ultra-taking-apple-silicon-to-a-new-extreme/
- https://en.wikipedia.org/wiki/Apple_M3
- https://support.apple.com/en-us/122211
- https://www.apple.com/newsroom/2025/03/apple-unveils-new-mac-studio-the-most-powerful-mac-ever/
- https://web.archive.org/web/20250312162945/https://www.apple.com/shop/buy-mac/mac-studio