GPT OSS 120B only has 5B active parameters. GP specifically said dense models, not MoE.