What happened
Tech Edition reports that Chuwi unveiled the UniBox AI495 Pro and lower-tier UniBox AI395 at IFA Berlin 2026. The AI495 Pro uses AMD’s Ryzen AI Max+ PRO 495 processor, up to 192GB of unified memory and integrated Radeon graphics. Chuwi says it can run models with up to 300 billion parameters using 4-bit quantisation. Pricing, release dates and independent performance results have not been disclosed.
Tech Edition reports that Chuwi introduced the UniBox AI495 Pro at IFA Berlin 2026 as a compact workstation for local AI processing. The system measures 209.6 x 209.6 x 67 millimetres and has an approximate 2.9-litre volume. The report says it includes a 2TB SSD, three M.2 PCIe x4 slots, two 10Gbps Ethernet ports, an OCuLink connection and two USB4 ports.
According to Tech Edition, the AI495 Pro pairs AMD’s Ryzen AI Max+ PRO 495, with 16 CPU cores, 32 threads and boost speeds up to 5.2GHz, with a 40-core Radeon 8065S GPU. The system has up to 192GB of LPDDR5x-8533 unified memory, with up to 160GB assignable as VRAM. Chuwi claims memory bandwidth of up to 273GB/s and up to 131 TOPS of combined AI performance, including up to 55 TOPS from the XDNA 2 neural processing unit.
Tech Edition says Chuwi claims the AI495 Pro can run models of up to 300 billion parameters with 4-bit quantisation. The outlet also reports Chuwi’s claim that the integrated graphics can outperform an Nvidia RTX 4060 in some 1080p gaming workloads. The source provides no independent benchmark results, test methodology or confirmation that these model-size and performance claims translate into useful real-world inference speeds.
Chuwi also announced the UniBox AI395, which Tech Edition describes as a lower-tier option using AMD’s Ryzen AI Max+ 395 processor. It supports up to 128GB of unified memory, with up to 96GB available as VRAM. Chuwi says it can run local models of up to 235 billion parameters with mixture-of-experts techniques and low-bit quantisation. Both systems reportedly include Windows 11 Pro, Wi-Fi 7, Bluetooth 5.4 and a 240W power adapter.
Why it matters
The reported launch targets a practical obstacle in local AI use: memory capacity. A workstation with a large shared CPU and GPU memory pool could let professionals and enthusiasts run larger models on a desk-sized system, potentially reducing reliance on cloud services for some workloads. That may be relevant for sensitive data, but the real value depends on software support, sustained performance, thermals, noise, power use and price. Tech Edition’s account relies substantially on Chuwi’s specifications and claims; those claims have not been independently confirmed in the source.
The reported configuration could make local inference of larger AI models more accessible in a small desktop form factor. Unified memory allows the CPU and integrated GPU to draw from the same pool, which is relevant when model weights would not fit comfortably within ordinary integrated-graphics memory.
Keeping workloads on the machine may be useful for organizations handling sensitive information, because prompts and data do not necessarily need to be sent to a remote cloud service. That is a potential practical benefit, not evidence that the systems provide complete privacy or security guarantees.
The source does not establish how quickly the machines run particular models, how long they can sustain peak performance, or whether their software ecosystem supports the claimed configurations. Chuwi’s specifications and comparisons therefore remain claims awaiting independent verification.
The absence of pricing and release dates makes it impossible to assess whether either workstation is competitive with other compact AI systems or conventional desktops.
What to watch next
The main unknowns are pricing, release timing, regional availability and independent testing. Buyers will also need evidence about which models can run at useful speeds, how much memory is actually available to applications, and whether the compact enclosure limits sustained workloads. Chuwi’s comparison with an Nvidia RTX 4060 is also presented as a company claim rather than an independently verified result.
Chuwi has not disclosed pricing or release dates for the UniBox AI495 Pro or UniBox AI395, and the source does not confirm markets or purchasing channels.
Independent tests should examine inference speed, memory allocation, thermals, fan noise, power consumption and performance over extended workloads.
Potential buyers should verify which operating systems, runtimes, quantisation formats and model architectures are supported before treating the stated parameter capacities as practical guarantees.