When my service sends bursts of requests, I’m trying to make its retry behavior respectful rather than just retrying on a fixed timer. Does the DeepSeek API include a reliable Retry-After value with 429 responses, or should clients use their own backoff strategy?

If the headers or limits differ between endpoints, examples from a production integration would be helpful. I’m especially interested in avoiding a retry loop that adds load while the account is still rate-limited.