The long thinking response shows multuple thinking elements, It should merge in one.

[Nishchit14]

The current “thinking” UI displays multiple elements, creating an unattractive interface. It should be consolidated into a single indicator, and after the response is complete, the indicator should remain for Thoufht for 15 seconds instead of showing a “thinking” text.

[TimFL]

You're potentially using the wrong component. Some LLMs have chain of thoughts, which return multiple isolated "thinking" entries / steps instead of having one that is constantly updated. Use the (newly added) chain of thoughts component instead.

[Nishchit14]

@TimFL It's not a chain of thought component, It's Reasoning component. I am testing it with gpt-5 model with high reasoning efforts.

[TimFL]

@TimFL It's not a chain of thought component, It's Reasoning component. I am testing it with gpt-5 model with high reasoning efforts.

What I mean is, some LLMs do not have a classic reasoning output but provide a list of thoughts (chain of thought) which ai-elements provides a new component for. I think GPT-5 has CoT instead of reasoning.

[Nishchit14]

@TimFL Thanks. I think you're right. but there should be the native handling for that within ai-element stack. The part.type is reasoning. which represents the reasoning block.

do you have any idea how to handle it?

[TimFL]

@TimFL Thanks. I think you're right. but there should be the native handling for that within ai-element stack. The part.type is reasoning. which represents the reasoning block.

do you have any idea how to handle it?

Late reply but I'd probably just render the single one for the first reasoning tag as a default, then when I detect a second one swap it out for the multi-step one assuming it's a chain of thought output then.