ukisai/Swift-Flash-Next

Swift 1.5 Flash Next

Chat demo for Swift 1.5 Qwen3.8-Flash-Next, UkisAI's reasoning-efficient fine-tune of Qwen3.8-Flash-Next. It uses 63.4% fewer thinking tokens while staying within 1 point of the base.

The model runs in NVFP4 on a UkisAI GPU server (4× H100, vLLM); this Space is the chat front end, with streaming reasoning and adjustable reasoning effort.

• —All Flash Next builds: Swift Flash Next collection
• —API: OpenAI-compatible at https://ukisai.com/api/flash-next/v1 (model flash-next), free for research