Qwen3.5-9B Hits 93.8% on HomeSec-Bench — 4.1 Points Behind GPT-5.4

SharpAI's HomeSec-Bench shows Qwen3.5-9B running locally on MacBook M5 Pro at 25 tokens/second using only 13.8GB unified memory achieves 93.8%, outperforming GPT-5.4-nano (92.7%) and nearly matching GPT-5.4-mini (95.8%). The larger Qwen3.5-35B-MoE delivers faster first-token response (435ms) than all tested OpenAI cloud models. Local models matching cloud on specialized tasks is now repeatable, not anecdotal.