[Bug] Error with "thinking or redacted_thinking" tokens not being processable with Sonnet 5.

Environment

• OS: Windows 11 Pro
• Eclipse Version: Version: 2026-06 (4.40.0) / Build id: 20260604-0653
• Plugin Version: GitHub Copilot GitHub Copilot 0.20.0.202607100822 com.microsoft.copilot.eclipse.feature

Describe the bug
While working on Sonnet 5, around the time i reach half of the max context (the context window icon shows half-circle full), i regularly get this message:

Sorry, an error occurred while generating a response. Details: unhandled status from server: 400 {"type":"error","error":{"type":"invalid_request_error","message":"messages.1.content.1: thinking or redacted_thinking blocks in the latest assistant messa... Read more from logs.

To Reproduce
This happens consistently on Sonnet 5 that i use often, around the time where the context seems above 50% of max size. I definitely did not see that with any other models, but i also do not use other models often enough. To "fix", i switch to Sonnet 4.6 to continue my task and that works, but i can never get back to Sonnet 5 short of a whole new session. This happens no matter the "thinking" mode (low/high) i have used. This happens frequently, but not always under the "50%" condition i noted. I cannot reproduce this with a fixed set of steps.

Expected behavior
A session using Sonnet 5 shouldn't die this way.

Screenshots
The first screenshot is of the errors, then the second shows after i switched to Sonnet 4.6 to continue.

Additional context
N/A

Start by inspecting the logs referenced in the server error and compare the failing Sonnet 5 session with the successful Sonnet 4.6 continuation. Determine where the thinking or redacted_thinking content becomes unprocessable; done means the session can continue with Sonnet 5 without the 400 error.