Apple · Apple M5 Ultra · 2026
Apple Mac Studio M5 Ultra 256GB for local AI
Most of what the M5 Ultra offers, with 256GB, shipping now at $9,499.
Updated · Speeds marked ~ are estimates; seemethodology
- Unified memory
- 256GB
- Bandwidth
- 1,228 GB/s
- Power
- 480W
- Runs well
- 14 models
M5 Ultra 256GB
Typically about $9,499
Can the M5 Ultra 256GB run LLMs?
The M5 Ultra 256GB has 256GB of unified memory (about 192GB usable by the GPU by default) and 1,228 GB/s of memory bandwidth, which decides how big a model fits and how fast it generates text. It comfortably runs models up to about 320B parameters: the largest in our catalog is GLM-5.3-Flash at Q3_K_M (161GB with 8K context, ~39 tokens/sec). Our top all-round pick for it is DeepSeek V4 Flash (FP4, ~45 tokens/sec).
Recommendations
Best models for the M5 Ultra 256GB
Best overall · also coding, reasoning
DeepSeek V4 FlashFP4 · 168GB · ~45 tokens/sec
Runs wellRunner-up · also vision
Qwen3.8 27BQ8_0 · 32GB · ~22 tokens/sec
Runs wellFastest good model
gpt-oss-20bMXFP4 · 15GB · ~105 tokens/sec
Runs wellLargest model that fits
GLM-5.3-FlashQ3_K_M · 161GB · ~39 tokens/sec
Runs well
Compatibility
What runs on the M5 Ultra 256GB
Every current model in our catalog, at the best quantization that fits in the ~192GB the GPU can use with 8K tokens of context. "Partial offload" means part of the model runs from system RAM (we assume 64GB), which is much slower.
| Model | Parameters | Best quant | Memory | Speed | Verdict |
|---|---|---|---|---|---|
| Gemma 4 E4B | 8B | Q8_0 | 10 GB | ~64 tok/s | Runs well |
| Qwen3.5 9B | 9B | Q8_0 | 12 GB | ~57 tok/s | Runs well |
| Phi-4 | 14.7B | Q8_0 | 18 GB | ~41 tok/s | Runs well |
| gpt-oss-20b | 20.9B (3.6B active) | MXFP4 | 15 GB | ~105 tok/s | Runs well |
| Gemma 4 26B A4B | 26B (3.8B active) | Q8_0 | 30 GB | ~73 tok/s | Runs well |
| Qwen3.8 27B | 27B | Q8_0 | 32 GB | ~22 tok/s | Runs well |
| Gemma 4 31B | 31B | Q8_0 | 36 GB | ~20 tok/s | Runs well |
| Qwen3.6 35B-A3B | 35B (3B active) | Q8_0 | 41 GB | ~83 tok/s | Runs well |
| GLM-4.5-Air | 106B (12B active) | Q8_0 | 118 GB | ~28 tok/s | Runs well |
| Llama 4 Scout | 109B (17B active) | Q8_0 | 120 GB | ~21 tok/s | Runs well |
| gpt-oss-120b | 117B (5.1B active) | MXFP4 | 68 GB | ~93 tok/s | Runs well |
| Qwen3.5 122B-A10B | 122B (10B active) | Q8_0 | 134 GB | ~33 tok/s | Runs well |
| DeepSeek V4 Flash | 284B (13B active) | FP4 | 168 GB | ~45 tok/s | Runs well |
| GLM-5.3-Flash | 320B (18B active) | Q3_K_M | 161 GB | ~39 tok/s | Runs well |
| GLM-5.3 | 744B (40B active) | — | — | — | Won't fitWhat it needs |
Performance
M5 Ultra 256GB vs. alternatives on Qwen3.8 27B
Generation speed in tokens per second for a single user. Token generation is limited mostly by memory bandwidth, which is why bandwidth predicts these results so well.
M5 Ultra 256GB
Q8_0 · 256GB
~22tok/s
M5 Ultra 512GB
Q8_0 · 512GB
~22tok/s
M5 Max 128GB
Q8_0 · 128GB
~13tok/s
Ryzen AI Max+ PRO 495
Q8_0 · 192GB
~6.9tok/s
Measured
M5 Ultra 256GB benchmark results
| Model | Quant | Runtime | Generation | Prompt | Source |
|---|---|---|---|---|---|
| Llama 2 7B Chat | Q4_0 | llama.cpp (Metal) | 180 tok/s | 4945 tok/s | llama.cpp discussion #4167: Performance of llama.cpp on Apple Silicon M-series · October 2026 |
| Llama 2 7B Chat | Q8_0 | llama.cpp (Metal) | 120 tok/s | 4910 tok/s | llama.cpp discussion #4167: Performance of llama.cpp on Apple Silicon M-series · October 2026 |
| Llama 2 7B Chat | F16 | llama.cpp (Metal) | 74 tok/s | 5205 tok/s | llama.cpp discussion #4167: Performance of llama.cpp on Apple Silicon M-series · October 2026 |
Our take
Should you buy the M5 Ultra 256GB for AI?
This configuration pairs the M5 Ultra with 256GB of unified memory and the same quoted 1.2TB/s of bandwidth as the 512GB model. About 192GB is available to the GPU by default, enough for 200B-class MoE models such as Qwen3-235B at 4 bit or 70B dense models at 8 bit with long context. Unlike the 512GB model, it was already shipping in October 2026, at $9,499. It costs much more than a 128GB M5 Max, but it doubles both the memory and the bandwidth.
What we like
- 256GB of fast unified memory, available now
- About 1.2TB/s of bandwidth
- Quiet and power efficient
Watch out for
- Cannot fit 600B-plus models at 4 bit
- No CUDA support
- Memory is fixed at purchase
Best for: Running 200B-class MoE models · Long-context work with 70B models
Specifications
M5 Ultra 256GB specs that matter for AI
- Memory
- 256GB LPDDR5X (unified)
- GPU-usable memory
- ~192GB by default
- Memory bandwidth
- 1,228 GB/s
- Architecture
- Apple M5 Ultra
- Power
- 480W (system)
- Interface
- Thunderbolt 5 (no PCIe slots)
- Released
- 2026
- Launch price
- $9,499
- Street price (new)
- ~$9,499 (2026-10)
- Software
- Metal, MLX
macOS lets the GPU use about 75 percent of memory by default (about 192GB), and the iogpu.wired_limit_mb setting can raise it. Apple quotes 1.2TB/s for M5 Ultra.
Sources: Apple Mac Studio (M5 Max or M5 Ultra) tech specs, MacRumors: Mac Studio with M5 Ultra and 512GB launching in October, ModelPiper: raising the Metal memory ceiling with iogpu.wired_limit_mb
Cross-shop
Alternatives to the M5 Ultra 256GB
Frequently asked questions
How much VRAM does the M5 Ultra 256GB have?
The Apple Mac Studio M5 Ultra 256GB has 256GB of unified memory (about 192GB usable by the GPU by default) with 1,228 GB/s of memory bandwidth.
What is the largest LLM the M5 Ultra 256GB can run?
Fully in GPU memory, the largest model in our catalog that runs comfortably is GLM-5.3-Flash (320B (18B active)) at Q3_K_M, using about 161GB with 8K context at ~39 tokens/sec.
What is the best LLM to run on the M5 Ultra 256GB?
Our top pick is DeepSeek V4 Flash at FP4, which fits in 256GB with room for 8K of context and runs at ~45 tokens/sec. See our full list of the best models for the M5 Ultra 256GB for coding, reasoning, and vision picks.
Is the M5 Ultra 256GB good for local AI in 2026?
Most of what the M5 Ultra offers, with 256GB, shipping now at $9,499.