AMD announced that its Ryzen AI Max+ Series processors can run large language models like GPT-OSS 120B entirely on local hardware, eliminating the need for cloud APIs. The processors deliver 1.7 times more tokens per dollar compared to NVIDIA DGX Spark in LM Studio testing, marking a significant shift in how developers and enterprises deploy AI workloads. Systems powered by Ryzen AI Max+ 392 and 388 will ship from Acer and ASUS starting Q1 2026.
What AMD Ryzen AI Max+ Delivers
The Ryzen AI Max+ Series enables users to run GPT-OSS 120B, a model with 116.8 billion total parameters—directly on systems with 128GB of unified memory. This eliminates dependency on remote servers or API rate limits.
AMD’s testing shows the processor runs GPT-OSS 120B approximately 10 times faster than Meta Llama 3 70B despite the larger parameter count. The performance advantage stems from more efficient model architectures with lower activated parameters and a dramatically expanded context window of 128K tokens compared to the previous 8K standard.
The new Ryzen AI Max+ 392 features 12 cores with boost speeds up to 5GHz, while the 388 offers eight cores, both including 50 TOPS NPUs and 60 TFLOPS GPU performance. All Ryzen AI Max+ chips support up to 128GB of LPDDR5X memory with 256GB/s bandwidth, the first x86 processors to offer this configuration.
Why Local AI Inference Matters Now
Running large models locally gives organizations control over sensitive data without sending information through third-party APIs. Network-independent performance means predictable response times regardless of connectivity issues.
Cost predictability is another major factor. Cloud API pricing varies based on token consumption, making budget forecasting difficult for enterprises deploying AI at scale. Local inference on Ryzen AI Max+ hardware converts variable operational expenses into fixed capital costs. AMD’s benchmark shows an average of 1.7x more tokens per dollar when comparing the Framework Desktop ($2,566) with Ryzen AI Max+ 395 against NVIDIA DGX Spark ($4,000) across four models: GPT-OSS 20B, GPT-OSS 120B, GLM 4.5 Air, and DeepSeek R1 Distill 70B.
Performance Across Windows and Linux
The Ryzen AI Max+ platform supports both Windows productivity tools and Linux-native machine learning frameworks without requiring separate systems.
Windows compatibility matters for CAD applications and creative suites that only run officially on Microsoft’s operating system. Developers can simultaneously access Linux environments for model training and deployment without dual-boot configurations or virtual machine overhead.
Real-world testing on Reddit’s LocalLLaMA community shows users achieving approximately 50 tokens per second with GPT-OSS 120B on the Ryzen AI Max+ 395 (128GB), dropping to around 30 tokens per second at 32,000-token context lengths while consuming roughly 120 watts.
Extended Context Windows Enable New Workflows
The 128K token context window in GPT-OSS 120B changes how professionals handle large documents. Contract review tools can process entire agreement sets in a single pass rather than breaking them into sections.
Coding assistants gain visibility into more repository code, logs, and documentation simultaneously. Analysts can input complete multi-year reports and time series data without manual chunking or complex prompt engineering strategies. Workflows that previously required careful document splitting now fit within a single model inference.
Market Positioning and Availability
AMD expanded the Ryzen AI Max lineup at CES 2026 with the 392 and 388 models joining the existing 395, 390, and 385 SKUs. The Max+ branding indicates a GPU-heavy configuration with fully-enabled 40 compute unit integrated graphics delivering 60 TFLOPS peak performance.
First systems ship in January 2026 with broader OEM availability throughout Q1 2026 from partners including Acer and ASUS. The Framework Desktop and ROG Flow Z13 already demonstrated the platform’s capabilities in small form factors without dedicated GPUs.
AMD CEO Lisa Su presented the Ryzen AI Max+ lineup alongside new Ryzen AI 400 Series mobile processors and Ryzen 7 9850X3D desktop chips during her CES 2026 keynote.
Featured Snippet Boxes
Can AMD Ryzen AI Max+ run ChatGPT-level models offline?
Yes. The Ryzen AI Max+ 395 with 128GB memory runs GPT-OSS 120B locally, achieving around 80% on the GPQA Diamond PhD-level science benchmark and 90% on MMLU college-level reasoning tests—comparable to ChatGPT quality without cloud APIs.
How much memory do you need for GPT-OSS 120B on AMD Ryzen AI Max+?
GPT-OSS 120B requires 128GB of unified memory when quantized to 4-bit precision. The AMD Ryzen AI Max+ 395 is the first x86 processor to support this memory configuration in a single system.
What is the cost difference between AMD Ryzen AI Max+ and NVIDIA for AI workloads?
AMD’s testing shows the Ryzen AI Max+ 395 in a Framework Desktop ($2,566) delivers 1.7 times more tokens per dollar than NVIDIA DGX Spark ($4,000) when running GPT-OSS 120B, GPT-OSS 20B, GLM 4.5 Air, and DeepSeek R1 Distill 70B in LM Studio.
When will AMD Ryzen AI Max+ systems be available to buy?
Systems with Ryzen AI Max+ 392 and 388 processors ship starting January 2026, with broader availability from OEM partners including Acer and ASUS throughout Q1 2026. The Framework Desktop with Ryzen AI Max+ 395 is already available.
