Weird responses with Llama 4 scout.

[danschind2]

I'm wondering if this is an issue with continue.dev or the model itself. On llama 3.1 70b when I ask the plugin to make a file, I get this.

Then when I test on llama 4 scout, the response looks like this.

Do you guys think this is the model or an issue with how the tool is called in continue.dev

[danschind2]

Anyone have any thoughts on this?

[BradKML]

When in doubt, switch models. we know that something is fishy in the way many companies are fine-tuning their models to use certain tool call formats (OpenAI Harmony, Anthropic API, Qwen3 Code)

[uinstinct]

Can you try turning on system message tools and check if it is still happening?