Code generation with Ollama (qwen3:14b) inserts model's think text

[GihoKwak]

I'm using the Continue extension with a locally served Ollama model (qwen3:14b) for code assistance.

Ever since the most recent update, I've been experiencing a issue: when I accept a code suggestion from the assistant (e.g., remove all comments from code), the LLM's "think" process text is being injected directly into my editor along with the code.
In some cases, this action also causes existing, unrelated code to be deleted.

It seems like the extension might be incorrectly parsing the model's full stream output, including metadata or the chain-of-thought, instead of just the final code block.

Has anyone else encountered this behavior?
Any insights on what might be causing this or a potential fix would be greatly appreciated.

[faetschi]

#5419

seems to be fixed, but not sure?

• IntelliJ v1.0.48
• VSCode v1.3.22

still have these issues with various tool capable models