Don't lose money on Z.ai's GLM Coding Plan — for any agent that touches images, it's useless. The endpoint is TEXT-only:...

Don't lose money on Z.ai's GLM Coding Plan — for any agent that touches images, it's useless. The endpoint is TEXT-only: one screenshot in your agent's history = HTTP 400 on every call. Your framework silently falls back to a pay-per-token vendor — you pay the flat plan AND the fallback bill, thinking you're on GLM.Irony: the model (GLM-5.3-Flash) is multimodal. The plan just doesn't carry images. Our vision session ran hours on DeepSeek unnoticed.Checks and fixes — new post:#AI #LLM #GLM