{"type": "article", "title": "Inference-only expert boost saves 8.5% reasoning tokens in Qwen 35B MoE", "publisher": "Web Pulse", "url": "https://wpnews.pro/news/inference-only-expert-boost-saves-8-5-reasoning-tokens-in-qwen-35b-moe", "original_source": "https://zenodo.org/records/22255483", "published": "2026-09-02T22:10:23+00:00", "accessed": "2026-09-03", "id": "inference-only-expert-boost-saves-8-5-reasoning-tokens-in-qwen-35b-moe"}