Open-weight & owned stack · Training tuning
UkisAI Releases Faster Swift-Qwen3.8-27B Model
UkisAI released Swift-Qwen3.8-27B, claiming nearly twice the inference speed with substantially fewer reasoning tokens and minimal accuracy loss.
Read the original at reddit.comOpens the publisher's site in a new tabAlso covering this
Cut Qwen3.8-27B Reasoning Tokens by 40% -- 3.8 'ThinkingCap' benchmarked!reddit.com, Sep 15Thank you :) Swift Qwen 3.8 27B now has 100k+ downloads, is #1 finetune and #9 model on HuggingFace Trendingreddit.com, Sep 17