DeepSeek Routes All V4 Pro API Traffic to V4.1 Flash, Cutting Costs by Over 70%

My take: Starting today, if your team uses the DeepSeek API with the V4 Pro model ID, your next bill will be at least 70% cheaper. DeepSeek routed all that traffic to the V4.1 Flash model with no code changes required. This is not an upcoming announcement: it has been live in production since 4:00 UTC on September 14.

What matters here is not just the price cut: it is the signal that inference costs continue to fall faster than most teams update their tool budgets. If your organization set a monthly spending cap on AI APIs six months ago, you are probably overpaying.

When did you last compare what you pay per AI model or provider, and check whether the cost has already dropped since then?