AnalysisAI ModelsNovember 10, 2026

Prototype runs 120B MoE models on 32GB M1 Max under 16 GiB

Read original source →mitln.github.io

A prototype engine ran gpt-oss-120b and Qwen3.5-122B-A10B with a 16 GiB process-memory budget by keeping MoE experts on SSD. On gpt-oss-120b it averaged 7.84 tokens/s vs 5.78 for TUFF 10.2.2, about 36% faster; under 4 GiB competing load it lost ~3% vs TUFF's ~23%.

1 source

More stories today

Open the live feed