DeepSeek V4 Lands Mid-July and Rewrites AI Pricing
China's DeepSeek has set a mid-July launch for the official V4, with a new API pricing model that piles fresh pressure on OpenAI and Anthropic.
- The official DeepSeek V4 arrives mid-July, with new optimizations and a reworked API pricing model.
- The preview dropped on April 24, with two open models: V4-Pro (1.6 trillion parameters) and V4-Flash (284 billion), both with a 1 million token context.
- The new pricing adds peak hours during which the API rate doubles.
- Cheap Chinese models have already pushed OpenAI and Anthropic to rethink their prices.
How much does it still cost to talk to an AI? From mid-July, with DeepSeek, the answer depends on the time of day.
The Chinese company says the official version of its V4 model lands mid-July. It brings new optimizations, but also a pricing scheme that changes the rules for anyone who treats AI as a tool rather than a toy.
What exactly does DeepSeek V4 bring?
Start with what is already known. The preview shipped on April 24 and delivered two open-weight models. The larger is V4-Pro with 1.6 trillion parameters, the smaller is V4-Flash with 284 billion. Both run on a one million token context, which means they can hold a long document or an entire project in their head at once.
For anyone who writes, codes, or analyzes, a million tokens is a concrete thing. You feed in a whole book, a full contract, or an entire code repository, and the model does not lose the thread halfway through. Until recently, that was the privilege of the most expensive Western models.
The official mid-July version should add further tuning over the preview. The company is not revealing every detail yet, but the message is clear: the preview was the warm-up, and now the real thing is coming.
Open weights are the key phrase here. They mean the model is not locked behind a single company. Firms and individuals can run it on their own hardware, without shipping data to someone’s cloud. For privacy and for cost control, that is a big difference.
Why peak pricing changes everything
Now for the part many people will dislike. For the first time, DeepSeek is introducing pricing by the hour. During peak demand, in the morning and afternoon, the cost of using the API doubles compared with quieter hours.
The logic is the same as with electricity or plane tickets. When everyone rushes in, you pay more. When the crowd thins out, you pay less. For heavy users processing millions of requests, this is not a rounding error, it is a budget line that demands planning.
There is a practical tip buried in all of this. If your job can wait, shifting heavier tasks to the cheaper hours cuts the bill directly. Automated reports, bulk data processing, and overnight jobs suddenly make sense beyond the pure engineering of it.
So what does this announcement mean for the big Western players? Cheap Chinese models have already pushed OpenAI and Anthropic to reconsider their prices. DeepSeek showed that a serious model can be offered far cheaper, and that changed everyone’s bargaining position.
AI prices fall when someone shows up doing the same work for less money. That is good news for anyone paying those bills, from a single developer in Belgrade to a large firm just now bringing AI into daily work. Competition does what it always does, it drags the price down.
Our take is simple: if you already use AI for work, watch what the official V4 offers against what you pay now. Peak pricing looks like a complication, but with a little planning it can save you money.
Would you shift part of your AI workload to cheaper hours to halve the bill, or does response speed matter more to you than cost?
