test-time compute

Open-source 27B model proves 'slow thinking' — test-time compute hits mid-tier

Reddit's local LLM community drops a Qwen-based 27B variant: slower per-token, better quality. First 'slow thinking' win for mid-sized open models.