DeepSeek, the Chinese lab that makes the cheapest serious AI on the market, is about to change how it charges for its newest model: more during the busy hours, the same the rest of the day.
What DeepSeek V4 is
DeepSeek is the open-source lab that shook the industry early last year, when its model matched far more expensive American AI for a tiny fraction of the cost and briefly rattled the US stock market. Because its models are open, they show up free or near-free inside other apps, which has quietly forced down what everyone else can charge.
DeepSeek V4 is its next flagship model. It has technically been available as a "preview" for about two months, but the full, official version is set to land in mid-July. People who have used the preview say it has been good but not yet great, so the real excitement is around how much sharper the finished release will be.
What "peak-hour pricing" actually means
Here is the new part. Today DeepSeek charges one flat rate to use its model. With V4, that changes: during the busiest stretches of the day in China (roughly 9am to noon and 2pm to 6pm Beijing time), the price will double. Outside those windows, the price stays exactly where it is now.
This is the same idea as Uber's surge pricing or peak electricity rates. When everyone wants the service at once, it costs more, which nudges people who can wait to run their work at quieter, cheaper times and takes some strain off DeepSeek's computers.
For people in the US, the timing is almost a gift: those Beijing peak hours fall in the American overnight, so most US daytime use stays in the cheap, off-peak window.
Still the cheapest by far
It is worth keeping the scale in mind. Even after the peak-hour doubling, DeepSeek's most expensive rate is still only a small fraction of what comparable American models cost. One widely shared comparison put DeepSeek's priciest tier at around a tenth of a US competitor's rate. As one user put it, the "almost-charity pricing" era had to end at some point, but the new prices are hardly expensive.
I feel like open source crossed a chasm this month with GLM 5.2, so I'm genuinely excited to see what the full DeepSeek V4 can do when it lands in July.
The pricing strategy is interesting too. Charging more only at peak hours is a smart way to handle a flood of demand, and those windows land at odd hours for most of the world anyway.
Sources
- @teortaxesTex — DeepSeek V4 coming mid-July with peak-hour pricing
- @BiteyeHQ — the peak/off-peak rate details
- @yuhasbeentaken — what the peak hours mean in US time
- @poezhao0605 — DeepSeek vs US pricing comparison
- @hqmank — "the almost-charity pricing era had to end"
- @jun_song — preview performance and the mid-July release