DeepSeek has announced the official version of its V4 models for API users, confirming a mid-July 2026 launch. The move covers both deepseek-v4-pro and deepseek-v4-flash, the same two models that have run in preview since April 24, 2026. This is confirmed by DeepSeek’s official V4 preview announcement, since the official launch itself has not happened yet and the AllBetaInfo team has not tested it firsthand. The announcement also introduces a new peak hour pricing system for the API.
What Changes When V4 Goes Official
DeepSeek V4 is not a new model lineup. deepseek-v4-pro and deepseek-v4-flash have already been public since April 24, 2026. The mid-July date marks their move from preview status to a full official release.
The official version keeps the same two models. DeepSeek says the update brings performance improvements and optimizations, without replacing either model.
How Peak Hour Pricing Works
The biggest change is pricing, not the models themselves. Starting with the official launch, DeepSeek will charge double the standard rate during two daily peak windows.
Peak hours run from 9 AM to 12 PM and 2 PM to 6 PM, Beijing time, every day. Off peak pricing stays the same as the current preview rate. DeepSeek says the goal is more reasonable resource allocation and better service stability.
DeepSeek will send a 24 hour email notice before any pricing change takes effect. Users who disagree with the new pricing can stop using the service and request a refund of their remaining balance under DeepSeek’s stated policy.
Why Beijing Time Matters for Global Developers
Beijing peak hours fall late at night and overnight for developers in the United States, roughly 9 PM to 6 AM Eastern time. This means many American developers running workloads during their normal business day would actually land in DeepSeek’s off peak pricing window.
Developers running interactive, user facing traffic during Chinese business hours will see the doubled rate. Teams running batch jobs, evaluation runs, or dataset generation can likely shift that work into off peak hours to avoid the higher cost.
What Happens to Older DeepSeek Models
The deepseek-chat and deepseek-reasoner model IDs will be fully retired on July 24, 2026. Both currently route to deepseek-v4-flash in non thinking or thinking mode.
Developers still calling deepseek-chat or deepseek-reasoner in production are already running on V4-Flash quality without a model ID update. DeepSeek recommends migrating to the new model IDs directly before the July 24 cutoff.
Platform Availability Right Now
The V4 preview models are available today through chat.deepseek.com and through the API. Both deepseek-v4-pro and deepseek-v4-flash support Expert Mode and Instant Mode on the web interface.
On the API side, both models support a 1 million token context window and integrate with OpenAI ChatCompletions and Anthropic style API formats. The official mid-July release does not change this access. It changes the pricing structure applied to API calls once the peak and off peak system takes effect.
Conclusion
DeepSeek confirmed the mid-July timeline for V4 going official on June 30, 2026, alongside its first ever time based pricing mechanism for a flagship API. The July 24, 2026 retirement of deepseek-chat and deepseek-reasoner falls just after the official launch, giving developers a narrow window to migrate model IDs before the older names stop working entirely.
Stay Updated with AllBetaInfo
Follow AllBetaInfo on Medium and X to catch every new feature under development and every update reaching beta testers first. AllBetaInfo is the global source for real-time beta updates on all major apps across Android, iOS, and Web. The team covers messaging, social, streaming, and AI apps, reporting each major change the moment it lands, so readers always stay a step ahead.