Swift-Qwen3.8-27B hits HF trending with 58% fewer thinking tokens

UkisAI's Swift-Qwen3.8-27B, a post-trained Qwen 3.8 27B targeting overthinking, tops Hugging Face trends with 58% fewer thinking tokens, doubled speed, and nearly unchanged accuracy.

2026-09-14 ~ 2026-09-14 · 2 related posts

• Swift-Qwen3.8-27b, a token-efficient reasoning Qwen finetune, trends on Hugging Face — ukisai · 2026-09-14
• Open-source Swift-Qwen3.8-27B cuts thinking tokens 58% for 1.95x speed at <1% accuracy loss — Secure_Recording_472 · 2026-09-14