A new bug has emerged in the last fortnight or so after using Post Assistant to connect open weight models to a local R installation.
It presents as the model responding to a message by indicating that it hasn't seen the content of the message. I had initally thought it was due to an agents.md or a audits.md file overloading the input token limit, but I'm able to replicate the fault in a new project, with a very rudimentary message:
Has anyone else observed this behavior? I'm using GLM 5.3 Flash hosted on Open Router