Tag
1 article
BottleCap AI releases ThinkingCap-Qwen3.8-27B, a fine-tuned version of Qwen3.8-27B that uses 37.2% fewer thinking tokens with only a minor accuracy trade-off.