Meet Swift.

Swift-Qwen3.8-27B is the first model in the UkisAI Swift series: reasoning models tuned to spend fewer tokens thinking before they answer.

View on Hugging Face Read the announcement

The difference

Shorter thinking.

17.7–46.2%

Fewer thinking tokens than the base model, depending on the task.

Drop-in replacement.

Same architecture, tokenizer, and chat template as the base model.

Architecture: Same
Tokenizer: Same
Chat template: Same

Available models

Swift · 27BSEP 9, 2026

Swift-Qwen3.8-27B

Shorter thinking. Same base-model interface.

View on Hugging Face

Start with Swift.

View on Hugging Face