AI: Agent sometimes displays unformatted <thinking> tags

[silvenon]

Summary

I haven't figured out a pattern when this happens, but occasionally the Agent displays unformatted <thinking> tags containing the LLM's thinking process, followed by the actual reply. It's confusing to see thinking text right next to what was intended for me.

Description

Steps to trigger the problem: I can't sorry. It doesn't even happen (only) with thinking models.

Actual Behavior: a subtle "thinking…" UI followed by the answer, where I could expand the thinking process if I wanted to. Expected Behavior: the <thinking> tags are not processed, the thinking process not intended for me is right next to the answer, I have to look for the closed </thinking> tag to know where to start reading.

Model Provider Details

• Provider: Anthropic via ZedPro
• Model Name: Claude Sonnet 4
• Mode: Agent Panel
• Profile: Ask

Zed Version and System Specs

Zed: v0.188.5 (Zed) OS: macOS 15.5.0 Memory: 16 GiB Architecture: aarch64

[max-frai]

Same for me, for some requests I see response with this raw thinking tag

[rtfeldman]

@silvenon @max-frai I haven't seen this in a long time - are you still seeing it on current Zed and Sonnet 4.5?

[max-frai]

Haven't seen last few weeks of daily usage

[rtfeldman]

Great! Let's close for now. 👍

[silvenon]

I still see it occasionally for other models, but for Sonnet 4.5 not that I recall.