DeepSeek has announced general availability of its V4-Pro model and a beta release of V4-Flash, alongside steep API price increases for the V4 family that take effect August 16 for most of the world. The hikes, which in some cases exceed 1,100%, come as the Chinese AI provider cites capacity constraints and pushes customers toward off-peak usage.
The new pricing schedule introduces peak and half-price off-peak rates for both models. DeepSeek says the structure aims to "allocate resources more reasonably" and encourage users to "schedule their tasks based on actual usage." Seventeen of every 24 hours fall into the off-peak window, meaning most buyers can still dodge the full increase.
What the new prices look like
For V4-Flash, the volume tier, off-peak input pricing with a cache miss is now $0.22 per million tokens, with output at $0.66 per million tokens. At peak, those figures rise to $0.44 and $1.32 respectively. The previous flat rates were $0.14 for input and $0.28 for output, putting the increase between 57% and 214% for input and between 136% and 371% for output.
V4-Pro, the premium tier, now costs $0.66 per million tokens for off-peak cache-miss input and $1.98 for output. Peak pricing runs $1.32 for input and $3.96 for output. Previously, Pro charged $0.435 for input and $0.87 for output, so the increase ranges from 51% to 203% on input and 127% to 355% on output.
Analysts see nuance behind the headlines
Mark Tauschek, VP of Research Fellowships and Distinguished Analyst at Info-Tech Research Group, cautioned against reading the raw numbers as a simple jump. "While it's alarming to see the headlines saying DeepSeek is raising API pricing by 50%-1100%, it doesn't really tell the whole story," he said.
Tauschek noted that the increase eliminates the price advantage that V4-Flash has over OpenAI 5.6 Luna at peak pricing, but not at off-peak pricing. It also does not erase V4-Pro's advantage over OpenAI GPT-5.6 Terra or GPT-5.6 Sol, even at peak. "This isn't unexpected at all; it's simple supply and demand," he added.
Sanchit Vir Gogia, Chief Analyst at Greyhound Research, agreed that context matters. "On paper, at peak, against the right comparator, DeepSeek's price advantage does disappear, and in places inverts," he said. "In practice, the schedule's own clock and cache hand most of it back to any buyer paying attention."
Gogia pointed to the cache as the weak spot. "The cache is where the advantage genuinely erodes," he said.
How DeepSeek compares with OpenAI
The pricing shift reshapes the competitive picture against OpenAI's Luna, which recently cut its off-peak API price by 80%. DeepSeek's measured cost per task remains about 60% below Luna even after that cut.
V4-Flash's off-peak edge over Luna has fallen from roughly 7x to about 3x. At peak, that advantage drops to 1.4x. On output alone, Flash off-peak is 45% cheaper than Luna, though input is marginally more expensive. V4-Pro at peak still runs about 5x cheaper than Luna on a representative coding-agent workload.
Gogia said the economics are shifting behavior. "Usage is following economics at least as much as capability, and economics can change by schedule," he said.
Stay ahead of the AI curve
The most important updates, news, and content — delivered weekly.
No spam. Unsubscribe anytime.
Who feels the pain
The new pricing hits DeepSeek's home market hardest and the export market lightest. Western buyers largely operate during off-peak hours, so their bills rise less. Tauschek said the impact will land mostly on developers. "Cost increases will mostly impact developers, but it will still be less expensive than most alternatives," he said.
Third-party providers have not yet reflected the price increase trend, but analysts expect they eventually will. Anthropic raised prices in April for the same reason: demand. DeepSeek's biggest selling point has always been its ultra-low price point, and demand for AI is increasing exponentially, with the company unable to keep up with compute requirements.
Many enterprises in the US do not use DeepSeek at all, and the company's pricing change does not alter the need for compatibility and multi-modality. Enterprises are adapting to model routing, which is critical for agentic workloads. A few months ago, organizations were paying per-seat pricing and running up usage as a matter of course. The market move to usage-based pricing has resulted in sticker shock akin to early cloud days.
The bigger picture
Gogia said CIOs should read the schedule with "relief and unease." Relief because off-peak rates soften the blow, unease because of what the pricing structure signals. "A supplier that has learned to price the clock has learned something about its own leverage," he said.
The deeper question, he argued, is substitutability. "The real question becomes whether lower economic floors, open weights, and compatible interfaces, when taken together with multi-model routing, make foundation model intelligence materially easier to substitute," Gogia said.
Open weights mean model developers become one of several parties able to serve inference requirements. The traditional software dependency changes shape when workloads can move between providers. "The vendor no longer owns the whole dependency once workloads can move between providers," Gogia said.
Tauschek said pricing will remain a central issue. "Pricing will continue to be a big deal because CFOs are starting to ask what they're getting for the massive AI spend," he said.
Gogia predicted that Flash will keep volume usage while Pro handles complexity, and that interface compatibility lowers the cost of adoption and departure. Capable inference, he said, can be produced far below the price structures that once surrounded frontier AI.
"The most lasting effect of DeepSeek is unlikely to be that it stayed cheapest," Gogia said. Instead, every provider must now explain why intelligence should command a premium once near-equivalent capability is available through several routes.
Both V4-Pro and V4-Flash support flexible reasoning levels, low, high, and max, along with thinking modes that use chain-of-thought reasoning. V4-Pro is available on app, web, and via API, and users can try it with "Expert Mode." V4-Flash remains in beta.
The price increases take effect August 16 for most of the world. DeepSeek's home market faces the steepest adjustments, while export customers, largely on off-peak schedules, see lighter increases. The company frames the change as a way to "allocate resources more reasonably" and push users toward "more flexible workload scheduling."