I’m testing JSON output with the DeepSeek API and wondering what happens when a response reaches its token limit before the object is complete. Does the API return a partial JSON string with a finish reason, or does it handle the incomplete output differently?

For anyone using the result in an application, how do you distinguish a complete response from a truncated one before trying to parse it? I’m interested in the behavior at the limit, rather than whether the model usually produces valid JSON on shorter prompts.