DeepSeek will release version 4 of its AI model in mid-July, introducing a time-based pricing structure that charges premium rates during business hours. The peak-valley mechanism reflects a growing trend among AI providers to manage compute costs and demand more effectively.

During peak hours—9:00 to 12:00 and 14:00 to 18:00 Beijing Time—the API costs double. The deepseek-v4-pro model charges ¥0.025 for cached input and ¥3.00 for uncached input per million tokens at standard rates, jumping to ¥0.05 and ¥6.00 respectively during high-demand periods.

For the lightweight deepseek-v4-flash variant, standard pricing sits at ¥0.02 for cached input and ¥1.00 for uncached input, with peak rates reaching ¥0.04 and ¥2.00. Output tokens cost ¥6.00 and ¥2.00 respectively under normal conditions, doubling during peak hours.

Users will receive email notifications 24 hours before any pricing changes take effect. This approach mirrors strategies used by major cloud providers and could influence how developers schedule compute-intensive workloads across different AI platforms.