If you pay for OpenCode Go, 12 of 15 models break on follow-ups. Here's why superior LLMs were having errors.

""Between two user messages, if the model performed a tool call, the intermediate assistant's reasoning_content must participate in the context concatenation and must be passed back to the API in all subsequent user interaction turns." api-docs.deepseek.com/guides/thinking_mode "If your code does not correctly pass back reasoning_content, the API will return a 400 error." That's not a provider bug. That's a client contract. The API explicitly tells you what to send, and if your client doesn't send