DeepSeek V4 Official Release Coming Mid-July With New Peak/Off-Peak API Pricing

What's new

• Official V4 release: V4-Pro and V4-Flash variants have been in early access since April 2026 (MIT license on HuggingFace); mid-July marks the full official deployment
• Peak/off-peak API pricing (first-ever time-of-day billing for DeepSeek):

- Peak hours: 9:00 AM–12:00 PM and 2:00 PM–6:00 PM Beijing time
- Peak rates = 2x off-peak rates
- V4-Pro example: off-peak output ¥6.00/MTok → peak ¥12.00/MTok

• 1-million-token context window across all V4 models
• Improved capabilities: stronger agent-based task execution, math reasoning, code generation
• 24-hour advance email notice before billing changes take effect

Why it matters

DeepSeek sparked the global AI price war in early 2026. Introducing surge pricing reverses that narrative and forces every AI API gateway and cost optimization tool to add time-of-day routing as a first-class dimension. For high-volume inference, routing batch jobs to off-peak slots (nights, weekends Beijing time) could yield 50% cost savings.

How to try it

• API: platform.deepseek.com — model IDs deepseek-v4-pro and deepseek-v4-flash
• Schedule batch jobs during off-peak hours (evenings/weekends in Beijing time)