GitHub radar
UkisAI Released the Free API of Vision Task for Swift-Qwen3.8-27B!
UkisAI has created a reasoning efficient version of Alibaba’s Qwen3.8-27B and released an OpenAI compatible API which supports text and image input of 262K tokens.
Swift-Qwen3.8-27B is a reasoning efficient version of Alibaba’s Qwen3.8-27B, developed by UkisAI. It supports both text and image inputs. With a context length of 262,144 tokens, after UkisAI discovered several tokens that may cause redundant reasoning, we fine-tuned this model to avoid using those tokens. So the new model has 58.3% fewer thinking tokens, 1.95x faster reasoning speed and less than 1% performance loss. UkisAI provides a free OpenAI compatible API service (ukisai.com) for individuals/companies with an annual income below 1 million USD.
Why a vibe-coder should care
This is a free solution when your agent needs to handle some image processing or document API processing tasks. Since it is OpenAI compatible, it is very convenient to integrate it with your current agent. But it is not designed for some complex reasoning tasks, just for some basic and repetitive tasks such as screenshot description, document page classification and image question answering.
How to install
Copy this and send it to your agent — Claude Code, Codex, any of them:
Connect this model's free API to my agent: https://huggingface.co/ukisai/Swift-Qwen3.8-27b — endpoint https://ukisai.com/api/swift/v1, OpenAI-compatible. Test a text query and a query with any image, show what the model responded.
This is a server-side model and cannot be run on a personal laptop. UkisAI provides a free API, so no local hardware is needed.
Open on Hugging Face▌ More finds