DeepSeek just cut its API cache prices to a tenth

DeepSeek has permanently slashed the price of cache-hitting input tokens across its entire API lineup, dropping the cost to one-tenth of the original rate for both V4-Pro and V4-Flash. The move makes these models dramatically cheaper for use cases where the same context is repeatedly accessed — and that's exactly what agentic coding looks like.

For V4-Flash, the cache hit price goes from $0.028 per million tokens to just $0.0028. Cache miss and output prices stay the same. For V4-Pro, DeepSeek is stacking this permanent cut on top of an existing 75% limited-time discount (running through May 5). The effective cache hit price for V4-Pro drops from yesterday's already-reduced $0.03625 down another order of magnitude to $0.003625 per million tokens.

The price cut matters most for agent workflows. When you're using a model for agentic coding — letting AI autonomously work through a codebase — the model repeatedly reads the same repository context over and over. Cache hit rates in those scenarios are far higher than in ordinary chat sessions, so the savings add up fast.

DeepSeek's V4 models already impressed with benchmark-topping coding performance, and at these prices, they're becoming an even more compelling option for developers building AI-powered coding tools.