Hey, this is a known issue with Composer 2 and Composer 2 Fast. It affects all CJK languages like Korean, Japanese, Thai, and more. The team is aware, and the Request ID has been accepted.
For now, the workaround is to use GPT-5.4 or Gemini 3.1 Pro for tasks with Korean text. These models handle UTF-8 correctly.
Related reports are collected here: [BUG]Korean characters broken (encoding issue) in Composer/Agent
There’s no ETA for a fix yet, but every report helps with prioritization. I’ll update the thread when I have more info.