I’m setting up a service that may send several DeepSeek API requests at once, and I’m trying to plan for rate limits before traffic grows. Are limits tied to an individual API key, the account, or the model being called?

If a burst gets throttled, what response details should a client inspect to decide when to retry? I’m also wondering whether using a different model changes the available quota or whether DeepSeek API rate limits are shared across the account.