The free desktop app uses an open-source engine to move model data from an SSD as needed, making some very large models usable on computers with far less memory than they typically require.
Running a large AI model locally usually means having enough memory to load most or all of that model. WiCi Infer takes a different approach: it reads the parts needed for each step from a computer’s SSD, then moves them into RAM or GPU memory as they’re needed.
WiCi, the startup behind the WiCi One wireless GPU, says its free desktop app can run open-weight models in the 800-billion-parameter class on Macs and PCs with as little as 16GB to 24GB of memory. The company recommends 48GB of memory and about 520GB of free SSD space.
That approach builds on InferSSD, WiCi’s open-source inference engine. The company’s broader idea is that some frontier-scale AI workloads could run on computers people already own, rather than requiring a large-memory workstation or a data center.
What the published benchmarks show
WiCi reports that a 48GB M5 Max MacBook Pro generated about 2.7 tokens per second using DeepSeek V4.1 Flash, a 763-billion-parameter model. The company says that was roughly three times the speed of standard llama.cpp on the same machine.
WiCi also reports about one token per second with Meta’s 816-billion-parameter Muse Spark Open. These are large models: according to WiCi, they would typically require more than 256GB of memory to run using conventional approaches.
Those figures are WiCi’s published results, and performance can vary with a computer’s hardware, model configuration, and workload. At the reported speeds, local use may be more suitable for experimentation or occasional tasks than for a fast, conversational assistant.
Local processing and a graphical interface
WiCi says Infer can connect to coding, writing, and document tools through a custom model endpoint. Because inference runs on the user’s computer, prompts and files can stay local rather than being sent to a cloud service. The company describes the app as having a graphical interface, so users can get started without working from the command line.
Local processing can be useful for people who want to keep sensitive material on their own machines. It also means users are responsible for their computer’s storage and performance, and running a model locally does not by itself verify the model’s accuracy or privacy settings in any connected software.
WiCi Infer is available for macOS and Windows. More details, including the company’s comparison with other local AI software, are available on WiCi’s product page. WiCi has also published a technical explainer on InferSSD.
Hence then, the article about wici infer aims to run 800b class ai models on everyday macs and pcs was published today ( ) and is available on MacSources ( Middle East ) The editorial team at PressBee has edited and verified it, and it may have been modified, fully republished, or quoted. You can read and follow the updates of this news or article from its original source.
Read More Details
Finally We wish PressBee provided you with enough information of ( WiCi Infer aims to run 800B-class AI models on everyday Macs and PCs )
Also on site :