ThinkingCap-Qwen3.6-27B: Qwen3.6 capabilities with 50% fewer thinking tokens · HackerLangs