DeepSeek Locks In 75% Price Cut on V4 Pro API After Promo Ends

DeepSeek is converting what had been a temporary 75% discount on its V4 Pro model API into the permanent list price. When the promotion ends on May 31, 2026, the post-promo rate will drop to one-quarter of the original launch price rather than reverting upward — effectively baking the discount in for good.

The pricing page also confirms a separate adjustment already in effect since late April: input cache hits across all models are now billed at one-tenth of the launch price, a meaningful reduction for workloads with repetitive prompt prefixes. DeepSeek is additionally retiring the deepseek-chat and deepseek-reasoner model names, folding them into the non-thinking and thinking modes of deepseek-v4-flash.

For developers, the move continues DeepSeek’s pattern of undercutting Western frontier-model pricing and pressuring competitors on cost-per-token. The cache-hit discount in particular favors agentic and long-context applications where prefix reuse dominates token spend.