Sometime overnight between 3 and 4 February 2026 (US time) a chatbot agent built using Copiliot Studio began returning "OpenAIModelTokenLimit" errors which were not being encountered before in pre-release testing or in production. These errors do not occur for all queries and can sometimes (but not always) be worked around by having the user clear their session state. We're not grounding in anything I'd think is too big or tricky (a few SharePoint sites, Confluence, one web site, and a few specific files).
Has anyone else encountered this situation in the last couple of days?
Update: as of 7 February 2026, like a fog in the morning sunlight, our errors have now evaporated and our agent is working as it did at the start of the week. My thanks to some unknown folks at Microsoft who silently fixed something that wasn't (officially) broken.
