AMD · RDNA 3.5 (Gorgon Halo) · 2026
AMD Ryzen AI Max+ PRO 495 (192GB) for local AI
192GB fits larger MoE models than Strix Halo, but bandwidth barely moves and system prices nearly doubled.
Updated · Speeds marked ~ are estimates; seemethodology
- Unified memory
- 192GB
- Bandwidth
- 273 GB/s
- Power
- 120W
- Runs well
- 12 models
Ryzen AI Max+ PRO 495
Typically about $6,800
Can the Ryzen AI Max+ PRO 495 run LLMs?
The Ryzen AI Max+ PRO 495 has 192GB of unified memory (about 144GB usable by the GPU by default) and 273 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 122B parameters: the largest in our catalog is Qwen3.5 122B-A10B at Q6_K (104GB with 8K context, ~14 tokens/sec). Our top all-round pick for it is Qwen3.6 35B-A3B (Q8_0, ~32 tokens/sec).
Recommendations
Best models for the Ryzen AI Max+ PRO 495
Best overall · also coding, reasoning, vision
Qwen3.6 35B-A3BQ8_0 · 41GB · ~32 tokens/sec
Runs wellRunner-up · also largest that fits
Qwen3.5 122B-A10BQ6_K · 104GB · ~14 tokens/sec
Runs wellFastest good model
gpt-oss-20bMXFP4 · 15GB · ~43 tokens/sec
Runs well
Compatibility
What runs on the Ryzen AI Max+ PRO 495
Every current model in our catalog, at the best quantization that fits in the ~144GB 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 | ~21 tok/s | Runs well |
| Qwen3.5 9B | 9B | Q8_0 | 12 GB | ~18 tok/s | Runs well |
| Phi-4 | 14.7B | Q8_0 | 18 GB | ~13 tok/s | Runs well |
| gpt-oss-20b | 20.9B (3.6B active) | MXFP4 | 15 GB | ~43 tok/s | Runs well |
| Gemma 4 26B A4B | 26B (3.8B active) | Q8_0 | 30 GB | ~27 tok/s | Runs well |
| Qwen3.8 27B | 27B | Q8_0 | 32 GB | ~6.9 tok/s | Runs well |
| Gemma 4 31B | 31B | Q8_0 | 36 GB | ~6.2 tok/s | Runs well |
| Qwen3.6 35B-A3B | 35B (3B active) | Q8_0 | 41 GB | ~32 tok/s | Runs well |
| GLM-4.5-Air | 106B (12B active) | Q8_0 | 118 GB | ~9.3 tok/s | Runs well |
| Llama 4 Scout | 109B (17B active) | Q8_0 | 120 GB | ~6.7 tok/s | Runs well |
| gpt-oss-120b | 117B (5.1B active) | MXFP4 | 68 GB | ~37 tok/s | Runs well |
| Qwen3.5 122B-A10B | 122B (10B active) | Q6_K | 104 GB | ~14 tok/s | Runs well |
| DeepSeek V4 Flash | 284B (13B active) | — | — | — | Won't fitWhat it needs |
| GLM-5.3-Flash | 320B (18B active) | — | — | — | Won't fitWhat it needs |
| GLM-5.3 | 744B (40B active) | — | — | — | Won't fitWhat it needs |
Performance
Ryzen AI Max+ PRO 495 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
Ryzen AI Max+ PRO 495
Q8_0 · 192GB
~6.9tok/s
DGX Spark
Q8_0 · 128GB
~6.9tok/s
Ryzen AI Max+ 395
Q8_0 · 128GB
~6.4tok/s
Our take
Should you buy the Ryzen AI Max+ PRO 495 for AI?
The Ryzen AI Max+ PRO 495, known as Gorgon Halo, refreshes Strix Halo with a Radeon 8065S GPU and support for 192GB of LPDDR5X-8533. On the same 256-bit bus that works out to 273 GB/s, only about 7 percent more than the Max+ 395. The extra memory fits larger MoE models such as Qwen3-235B at around 3 bit, but generation speed stays close to Strix Halo. Systems start at $6,799 for the Framework Desktop, which ships in November, and $7,399 for the Minisforum MS-S1 MAX-P495.
What we like
- 192GB of unified memory in a small box
- Runs larger MoE models than any Strix Halo system
- A full x86 PC with ROCm and Vulkan support
Watch out for
- 273 GB/s limits generation speed
- Systems cost about $6,800 or more
- Maximum GPU memory allocation not yet documented
Best for: Running 200B-class MoE models at home · Developers who need more memory than Strix Halo offers
Specifications
Ryzen AI Max+ PRO 495 specs that matter for AI
- Memory
- 192GB LPDDR5X-8533 (unified)
- Memory bandwidth
- 273 GB/s
- Memory bus
- 256-bit
- Architecture
- RDNA 3.5 (Gorgon Halo)
- Power
- 120W (system)
- Interface
- Integrated (mini PCs)
- Released
- 2026
- Street price (new)
- ~$6,800 (2026-10)
- Software
- ROCm, Vulkan
AMD has not documented the maximum GPU allocation for 192GB systems. On 128GB Strix Halo systems the limit was 96GB on Windows and about 120GB on Linux.
Sources: Phoronix: Framework Desktop with Ryzen AI Max+ PRO 495 pricing
Cross-shop
Alternatives to the Ryzen AI Max+ PRO 495
Frequently asked questions
How much VRAM does the Ryzen AI Max+ PRO 495 have?
The AMD Ryzen AI Max+ PRO 495 (192GB) has 192GB of unified memory (about 144GB usable by the GPU by default) with 273 GB/s of memory bandwidth on a 256-bit bus.
What is the largest LLM the Ryzen AI Max+ PRO 495 can run?
Fully in GPU memory, the largest model in our catalog that runs comfortably is Qwen3.5 122B-A10B (122B (10B active)) at Q6_K, using about 104GB with 8K context at ~14 tokens/sec.
Can the Ryzen AI Max+ PRO 495 run Llama 3.3 70B?
Yes. At Q8_0 it needs about 81GB with 8K context and runs at ~2.7 tokens/sec.
What is the best LLM to run on the Ryzen AI Max+ PRO 495?
Our top pick is Qwen3.6 35B-A3B at Q8_0, which fits in 192GB with room for 8K of context and runs at ~32 tokens/sec. See our full list of the best models for the Ryzen AI Max+ PRO 495 for coding, reasoning, and vision picks.
Is the Ryzen AI Max+ PRO 495 good for local AI in 2026?
192GB fits larger MoE models than Strix Halo, but bandwidth barely moves and system prices nearly doubled.