Unified memory · AMD
What AI models can a Ryzen AI Max+ 395 128GB run?
128 GB is server territory. The constraint stops being memory and becomes bandwidth: everything fits, and what varies is how long you wait for it. Memory is shared with the CPU, so the practical ceiling is lower than the sticker number: macOS hands roughly three quarters of it to the GPU by default.
A mini PC with 128 GB of shared memory at 256 GB/s, running Windows or Linux. It loads the same very large models as a Mac with the same capacity for less money, and answers at about half the speed of an M4 Max. Software support is improving and still behind Apple and NVIDIA.
The short answer
Assuming an 8k context window and default settings, these are the models worth downloading first.
See how fast it feels
Ryzen AI Max+ 395 128GB running GPT OSS 120B at MXFP4
YouWhy does my model use more memory when the conversation gets longer?
Reading your question
Simulated from our estimate at a 512-token question, not a recording. Assumes nothing else is competing for the GPU.
Where it sits
The GPU can use about 96 GB of this machine's 128 GB by default, so for planning it behaves like a 48 GB card with different bandwidth. The 48 GB guide explains what that tier runs and where it runs out, and the bandwidth guide explains why 256 GB/s decides how fast it answers.
Our indicative price for this machine is $2,199 new, reviewed 2026-09-01. That is 5.8 GB of memory and 12 GB/s of bandwidth for every $100. Prices move, so treat these as a place to start a search and not a quote.
Devices in a similar price band, fastest memory first:
| Device | Price | Memory | Bandwidth | Best all-round pick |
|---|---|---|---|---|
| Apple M1 Ultra 64GB | $2,400 | 64 GB | 800 GB/s | Qwen3.5 27B, 20.2 t/s |
| Apple M1 Max 64GB | $1,900 | 64 GB | 400 GB/s | Qwen3.5 35B A3B, 51.8 t/s |
| Apple M2 Max 32GB | $2,499 | 32 GB | 400 GB/s | Gemma 4 12B Instruct, 20.7 t/s |
| Apple M4 Pro 48GB | $1,799 | 48 GB | 273 GB/s | Gemma 4 26B A4B Instruct, 23.8 t/s |
Can a Ryzen AI Max+ 395 128GB run it?
A direct answer for the models people ask about most, each with the settings we would use.
Every model, scored on this device
Each row uses the highest-quality quantization that both fits and stays conversational. Speed is a single-stream estimate at 8k context.
| Model | Params | Verdict | Download | Memory used | Speed | Max context |
|---|---|---|---|---|---|---|
| GPT OSS 120B GPT-OSS | 116.83B (5.7B active) | Runs great | MXFP4 | 59.68 GB | 26.8 t/s | 128k |
| Qwen3 Coder Next Qwen | 79.67B (3.19B active) | Runs well | Q8_0 | 79.43 GB | 24.8 t/s | 256k |
| Qwen3 Next 80B A3B Instruct Qwen | 81.32B (3.19B active) | Runs well | Q8_0 | 81.07 GB | 24.8 t/s | 256k |
| Qwen3.5 35B A3B Qwen | 35.95B (2.9B active) | Runs great | Q8_0 | 36.13 GB | 27.7 t/s | 256k |
| GLM 4.7 Flash GLM | 31.22B (3.66B active) | Runs well | Q8_0 | 31.64 GB | 21.1 t/s | 128k |
| Gemma 4 26B A4B Instruct Gemma | 25.81B (3.74B active) | Runs well | Q8_0 | 26.72 GB | 18.7 t/s | 256k |
| Qwen3 30B A3B Qwen | 30.53B (3.34B active) | Runs well | Q8_0 | 31.3 GB | 21 t/s | 32k |
| GPT OSS 20B GPT-OSS | 20.91B (4.18B active) | Runs great | MXFP4 | 11.84 GB | 35.3 t/s | 128k |
| Qwen3 Coder 30B A3B Instruct Qwen | 30.53B (3.34B active) | Runs well | Q8_0 | 31.3 GB | 21 t/s | 32k |
| GLM 4.5 Air GLM | 110.47B (13.4B active) | Runs well | IQ4_XS | 56.5 GB | 10.5 t/s | 128k |
| Qwen3.5 9B Qwen | 9.65B | Runs well | Q8_0 | 10.25 GB | 16.5 t/s | 256k |
| Qwen3.5 27B Qwen | 27.78B | Runs well | IQ4_XS | 14.83 GB | 10.6 t/s | 256k |
| Gemma 4 E4B Instruct Gemma | 8B | Runs well | Q8_0 | 8.42 GB | 18 t/s | 128k |
| Phi 3.5 MoE Instruct Phi | 41.87B (6.64B active) | Runs well | Q8_0 | 42.84 GB | 11.2 t/s | 128k |
| Qwen3 14B Qwen | 14.77B | Runs well | Q6_K | 12.97 GB | 12.2 t/s | 32k |
| Qwen2.5 VL 7B Instruct Qwen | 8.29B | Runs well | Q8_0 | 9.03 GB | 18 t/s | 64k |
| Gemma 4 12B Instruct Gemma | 11.96B | Runs well | Q8_0 | 13.55 GB | 11.1 t/s | 256k |
| Qwen3 8B Qwen | 8.19B | Runs well | Q8_0 | 9.64 GB | 16.9 t/s | 32k |
| Qwen2.5 7B Instruct Qwen | 7.62B | Runs well | Q8_0 | 8.37 GB | 19.6 t/s | 32k |
| Qwen2.5 14B Instruct Qwen | 14.77B | Runs well | Q6_K | 13.22 GB | 11.9 t/s | 32k |
| DeepSeek Coder V2 Lite Instruct DeepSeek | 15.71B (2.74B active) | Runs great | Q8_0 | 16.12 GB | 29 t/s | 128k |
| Qwen2.5 Coder 14B Instruct Qwen | 14.77B | Runs well | Q6_K | 13.22 GB | 11.9 t/s | 32k |
| Granite 4.1 8B Granite | 8.79B | Runs well | Q8_0 | 10.35 GB | 14.6 t/s | 128k |
| Gemma 4 31B Instruct Gemma | 31.27B | Fits, but slow | Q8_0 | 34.67 GB | 4.2 t/s | 128k |
| Qwen2.5 Coder 7B Instruct Qwen | 7.62B | Runs well | Q8_0 | 8.37 GB | 19.6 t/s | 32k |
| Llama 3.1 8B Instruct Llama | 8.03B | Runs well | Q8_0 | 9.35 GB | 17.3 t/s | 128k |
| Phi 4 Phi | 14.66B | Runs well | Q6_K | 13.2 GB | 11.8 t/s | 16k |
| Mixtral 8x7B Instruct V0.1 Mistral | 46.7B (12.88B active) | Runs well | Q4_K_M | 27.66 GB | 10.3 t/s | 32k |
| Gemma 3 12B Instruct Gemma | 12.19B | Runs well | Q8_0 | 13.27 GB | 11.3 t/s | 128k |
| Qwen3 32B Qwen | 32.76B | Fits, but slow | Q8_0 | 34.86 GB | 4.3 t/s | 32k |
Show the other 56 models
| Gemma 4 E2B Instruct | 5.12B | Runs great | Q8_0 | 5.44 GB | 28.4 t/s | 128k |
| Mistral 7B Instruct V0.3 | 7.25B | Runs well | Q8_0 | 8.58 GB | 18.1 t/s | 32k |
| Qwen3.5 4B | 4.66B | Runs great | Q8_0 | 5.27 GB | 29.6 t/s | 256k |
| Granite 4.1 30B | 28.87B | Fits, but slow | Q8_0 | 30.97 GB | 4.8 t/s | 128k |
| Mistral Small 24B Instruct 2501 | 23.57B | Runs well | Q4_K_M | 14.94 GB | 10.3 t/s | 32k |
| Internlm3 8B Instruct | 8.8B | Runs well | Q8_0 | 9.49 GB | 17 t/s | 32k |
| Mistral Nemo Instruct 2407 | 12.25B | Runs well | Q8_0 | 13.81 GB | 11.4 t/s | 128k |
| OLMo 3 7B Instruct | 7.3B | Runs well | Q8_0 | 10.13 GB | 15.6 t/s | 64k |
| Gemma 3 27B Instruct | 27.43B | Runs well | Q3_K_M | 14.02 GB | 10.7 t/s | 128k |
| Qwen2.5 32B Instruct | 32.76B | Fits, but slow | Q8_0 | 34.86 GB | 4.3 t/s | 32k |
| Mistral 7B Instruct V0.2 | 7.24B | Runs well | Q8_0 | 8.57 GB | 18.1 t/s | 32k |
| Qwen2.5 Coder 32B Instruct | 32.76B | Fits, but slow | Q8_0 | 34.86 GB | 4.3 t/s | 32k |
| Codestral 22B V0.1 | 22.25B | Runs well | Q4_K_M | 14.73 GB | 10.3 t/s | 32k |
| Gemma 2 9B Instruct | 9.24B | Runs well | Q8_0 | 11.5 GB | 13.1 t/s | 8k |
| LFM2.5 8B A1B | 8.47B | Runs great | Q8_0 | 8.81 GB | 51.2 t/s | 64k |
| Qwen3 4B | 4.02B | Runs great | Q8_0 | 5.46 GB | 28.5 t/s | 32k |
| Gemma 3 4B Instruct | 4.3B | Runs great | Q8_0 | 4.91 GB | 32 t/s | 128k |
| DeepSeek Coder 7B Instruct V1.5 | 6.91B | Runs well | Q8_0 | 10.99 GB | 14.3 t/s | 4k |
| Llama 3.3 70B Instruct | 70.55B | Fits, but slow | Q8_0 | 72.85 GB | 2 t/s | 64k |
| Qwen2.5 VL 72B Instruct | 73.41B | Fits, but slow | Q8_0 | 75.68 GB | 2 t/s | 64k |
| Qwen2.5 72B Instruct | 72.71B | Fits, but slow | Q8_0 | 74.99 GB | 2 t/s | 32k |
| DeepSeek Coder 6.7B Instruct | 6.74B | Runs well | Q8_0 | 11.07 GB | 13.8 t/s | 16k |
| Phi 4 Mini Instruct | 3.84B | Runs great | Q8_0 | 5.17 GB | 30.3 t/s | 128k |
| Llama 3 3 Nemotron Super 49B V1 | 49.87B | Fits, but slow | Q8_0 | 69.89 GB | 2.1 t/s | 16k |
| Granite 4.1 3B | 3.4B | Runs great | Q8_0 | 4.34 GB | 36.5 t/s | 128k |
| Qwen2.5 3B Instruct | 3.09B | Runs great | Q8_0 | 3.68 GB | 43.6 t/s | 32k |
| Phi 3 Mini 4k Instruct | 3.82B | Runs great | Q8_0 | 4.9 GB | 32.8 t/s | 4k |
| Llama 3.2 3B Instruct | 3.21B | Runs great | Q8_0 | 4.42 GB | 35.9 t/s | 128k |
| LFM2.5 2.6B | 2.7B | Runs great | Q8_0 | 3.14 GB | 52 t/s | 128k |
| Qwen3.5 2B | 2.27B | Runs great | Q8_0 | 2.7 GB | 61.7 t/s | 256k |
| GLM 4.5 | 358.34B | Fits only at 2 to 3 bits | IQ2_XXS | 89.25 GB | 7.7 t/s | 16k |
| Gemma 2 2B Instruct | 2.61B | Runs great | Q8_0 | 3.54 GB | 45.6 t/s | 8k |
| Starcoder2 3B | 3.03B | Runs great | Q8_0 | 3.49 GB | 49 t/s | 16k |
| Qwen3 1.7B | 2.03B | Runs great | Q8_0 | 3.22 GB | 50.4 t/s | 32k |
| DeepSeek R1 Distill Qwen 1.5B | 1.78B | Runs great | Q8_0 | 2.3 GB | 83.1 t/s | 128k |
| Qwen2.5 1.5B Instruct | 1.54B | Runs great | Q8_0 | 2.06 GB | 83.5 t/s | 32k |
| SmolLM2 1.7B | 1.71B | Runs great | Q8_0 | 3.53 GB | 45.6 t/s | 8k |
| OLMo 2 0425 1B | 1.48B | Runs great | Q8_0 | 2.8 GB | 64.3 t/s | 4k |
| Llama 3.2 1B Instruct | 1.24B | Runs great | Q8_0 | 1.82 GB | 98.5 t/s | 128k |
| LFM2.5 1.2B Instruct | 1.17B | Runs great | Q8_0 | 1.59 GB | 130 t/s | 64k |
| Gemma 3 1B Instruct | 1B | Runs great | Q8_0 | 1.35 GB | 140 t/s | 32k |
| Qwen3.5 0.8B | 0.87B | Runs great | Q8_0 | 1.28 GB | 149.5 t/s | 256k |
| Qwen3 0.6B | 0.75B | Runs great | Q8_0 | 1.92 GB | 89.9 t/s | 32k |
| Qwen2.5 0.5B Instruct | 0.49B | Runs great | Q8_0 | 0.88 GB | 251.3 t/s | 32k |
| LFM2.5 350M | 0.35B | Runs great | Q8_0 | 0.74 GB | 389.2 t/s | 64k |
| Qwen3.5 122B A10B | 125.09B | Just fits | n/a | 96.23 GB | 12.9 t/s | n/a |
| MiniMax M3 | 427.04B | Fits only at 2 to 3 bits | n/a | 103.82 GB | 11.8 t/s | n/a |
| MiniMax M2.5 | 228.7B | Just fits | n/a | 115.46 GB | 11.5 t/s | n/a |
| MiniMax M2.7 | 228.69B | Just fits | n/a | 115.46 GB | 11.5 t/s | n/a |
| Qwen3 235B A22B | 235.09B | Just fits | n/a | 108.88 GB | 7.3 t/s | n/a |
| Kimi K2.6 | 1026.88B | Will not run | n/a | 578.44 GB | n/a | n/a |
| Kimi K2 Instruct | 1026.41B | Will not run | n/a | 578.18 GB | n/a | n/a |
| GLM 5.1 | 753.86B | Will not run | n/a | 425.04 GB | n/a | n/a |
| GLM 5.2 | 753.33B | Will not run | n/a | 424.75 GB | n/a | n/a |
| DeepSeek V3.2 | 685.4B | Will not run | n/a | 386.43 GB | n/a | n/a |
| DeepSeek R1 | 684.53B | Will not run | n/a | 385.95 GB | n/a | n/a |
What it will not run
11 of the 86 architectures we track are out of reach here, even at 2-bit precision.
Other devices with 128 GB
Same capacity, different speed. Once a model fits, bandwidth is what separates these.
Questions people ask about a Ryzen AI Max+ 395 128GB
How much VRAM does a Ryzen AI Max+ 395 128GB have?
The Ryzen AI Max+ 395 128GB has 128 GB of unified memory shared with the CPU, of which roughly 96 GB is available to a model once the operating system takes its share.
Can a Ryzen AI Max+ 395 128GB run a 70B model?
Yes. At Q8_0 a 70B model fits in 79.43 GB and generates around 24.8 tokens per second.
What is the best local model for a Ryzen AI Max+ 395 128GB?
GPT OSS 120B at MXFP4 is the strongest all-round fit: 59.68 GB used, around 26.8 tokens per second at 8k context.