Android Demo Runs 120B-Class MoE

BigMoeOnEdge shows that a OnePlus 15R with about 11GB of free memory can run oversized models on Android using only CPU and flash by loading only the active MoE experts on demand. The demo includes GPT-OSS-120B as well as Qwen 30B and Gemma 26B.

2026-07-18 ~ 2026-07-19 · 2 related posts

• Streaming 120B MoE Inference on Android — dai_app · 2026-07-18
• Running Ultra-Large MoE Models on Android — dai_app · 2026-07-19