According to the official announcement from DeepSeek, the new pricing for the DeepSeek-V4 series API took effect at midnight today. At the same time, V4-Pro has completed its testing phase and is now officially available for commercial use. This price adjustment abolished the previous unified billing model, adopting instead an industry-rare peak-valley tiered pricing system, with prices doubling during peak hours compared to off-peak hours.

Peak and Off-Peak Hours and Price Details

The official has designated 9:00-12:00 and 14:00-18:00 as peak hours daily, while other times are considered off-peak. The pricing during off-peak hours is only half of that during peak hours, with the billing unit uniformly set at yuan per million tokens. For example, for V4-Flash, the input cache hit cost is 0.05 yuan during off-peak hours, 1.5 yuan for a cache miss, and 4.5 yuan for output. During peak hours, these costs rise to 0.10 yuan, 3.0 yuan, and 9.0 yuan, respectively.

V4-Pro's pricing is generally one level higher. During off-peak hours, the input cache hit cost is 0.15 yuan, 4.5 yuan for a cache miss, and 13.5 yuan for output, while during peak hours, they are 0.30 yuan, 9.0 yuan, and 27.0 yuan, respectively. Both models support tool calls, Responses API, Anthropic-compatible interfaces, dialogue continuation, and FIM completion, with a maximum context output length of 384K.

The Computational Logic Behind Peak-Valley Scheduling

The core purpose of implementing peak-valley pricing is to optimize the allocation of computing resources and guide developers to schedule tasks during off-peak hours. Through market-oriented pricing levers, the platform aims to reduce computing load during peak hours and alleviate service instability caused by sudden increases in requests, thereby improving overall service availability.

Both V4 models are inherently adapted for long-text parsing, code development, intelligent agent deployment, and enterprise batch inference scenarios. The ultra-long 384K context can cover most production needs. For developers, scheduling non-real-time tasks during off-peak hours can significantly reduce call costs without sacrificing performance, marking a new stage in DeepSeek's transition from low-cost market capture to refined operations.