Developer Runs 120B LLM Across Phone and Five Computers

Wccftech reports that an individual ran a 120B large language model using an Android phone, three Macs and two Windows PCs. The source identifies the model as gpt-oss-120b.

The setup combined several lower-end devices rather than relying on a single machine with large amounts of memory. Wccftech says a 120B model would need at least 60GB to 80GB of combined video memory (VRAM) and system memory for decent performance, typically requiring expensive hardware.

The experiment came with a significant trade-off: Wccftech describes the distributed setup’s speed as extremely slow and says it is impractical. The report does not provide measured performance figures or further technical details about how the devices shared the workload.