Weird responses from GPT OSS models with latest release

[CalculonPrime]

I've had GPT OSS 20B F16 installed for a while, but after updating the Web UI to the latest release (haven't updated for a while) I noticed that the GPT OSS models all now start responding with:

analysis[thinking]

with no space after analysis. I have the "Enable thinking" checkbox unchecked.

For example, if I ask "Who was Khaddafi?" I get this:

analysisWe need to produce the continuation of the conversation from AI perspective. The AI needs to respond to the user's question ""Who was Khaddafi?"" presumably referring to Muammar Gaddafi, former Libyan dictator. Provide an answer. Should be helpful. The conversation style: previous AI responded with ""How can I help you today?"" The user asks question. We reply with answer. We might provide brief biography. ...

It does continue on to give an answer, but this initial part is weird, has formatting errors, and I don't remember it being present before. Seems to affect all GPT OSS models now.

Anyone else seeing this?