DeepSeek 4.1 Flash debugs some issues with GLM 4.7 responses and calls some of their instructions degenerate in their reasoning.

Degenerate instruction in question: "reply with just the
output" while testing reasoning streaming https://t.co/esUKBsjOSC