BottleCap AI Releases ThinkingCap-Qwen3.8-27B: 37.2% Fewer Thinking Tokens at a 0.86pp Accuracy Cost

BottleCap AI has released ThinkingCap-Qwen3.8-27B, a fine-tuned version of the Qwen3.8-27B model that reduces thinking token usage by 37.2% across 12 benchmarks. While the model experiences a slight 0.86 percentage point decrease in macro accuracy, it achieves a 2.25 percentage point improvement in long-context tasks and remains compatible with standard deployment frameworks like vLLM and SGLang.