DeepSeek quadruples AI model prices as discounted era ends
The Chinese AI firm’s V4 Pro and V4 Flash models see significant hikes, though the company maintains its offerings remain cheaper than key Western competitors.
DeepSeek has announced a fourfold increase in API pricing for its V4 models, marking a decisive shift away from the aggressive discounting strategy that helped establish its market position. The new pricing structure, which introduces peak and off-peak rates to allocate resources more reasonably, takes effect on 16 August.
Under the revised schedule, DeepSeek V4 Pro will cost $3.96 per million output tokens during peak hours, a substantial rise from the current rate of $0.87. Off-peak rates for the same model will be set at $1.98. Similarly, DeepSeek V4 Flash will see its peak-hour pricing increase from $0.28 to $1.32 per million tokens, with off-peak rates fixed at $0.66.
The move effectively reverses a commitment made in May, when the company announced that its initial discounted pricing—originally scheduled to expire on 31 May—would be made permanent. Despite this reversal, DeepSeek stated that the new structure is necessary to manage demand more efficiently as usage scales.
Even with the hikes, the company maintains that its models remain more affordable than many Western competitors. For context, Moonshot’s Kimi K3 model is priced at $15 per million output tokens, while OpenAI’s GPT-5.6 Sol costs $30. However, OpenAI’s lower-cost offering, GPT-5.6 Luna, is priced at $1.20, which is still higher than DeepSeek’s peak-hour Flash rate.
The price adjustment signals the end of the "cheap ride" that characterised DeepSeek’s early entry into the global AI market. While the firm previously leveraged significantly lower costs to challenge established players, the new tiered pricing suggests a maturation of its commercial strategy and a focus on sustainable resource management.
