DeepSeek Cashes In: V4 Pro Prices Jump as Peak Rates Arrive

The low-price playbook has changed

DeepSeek spent much of 2026 turning price into a competitive weapon. V4 Pro arrived with a one-million-token context window, serious coding and reasoning credentials, and a first-party API price low enough to force every competing model provider into an uncomfortable comparison. That strategy worked: developers adopted the model, infrastructure demand grew, and DeepSeek became one of the names that procurement teams used to challenge more expensive AI bills.

Now the company is cashing in on that success. The official V4 Pro rate card no longer offers one predictable price throughout the day. It divides weekdays into peak and off-peak periods, with peak usage costing twice the off-peak rate. More importantly, even the lower off-peak price is substantially above the previous flat rate in every V4 Pro token category.

How much more V4 Pro costs

Before the change, V4 Pro cost $0.003625 per million cache-hit input tokens, $0.435 per million cache-miss input tokens, and $0.87 per million output tokens. The current off-peak rates are $0.022, $0.66, and $1.98 respectively. That means the cheapest available period still raises cache-hit input by roughly 507%, cache-miss input by about 52%, and output by about 128%.

During peak hours, V4 Pro rises to $0.044 per million cache-hit input tokens, $1.32 per million cache-miss input tokens, and $3.96 per million output tokens. Compared with the old flat card, those changes are approximately 1,114%, 203%, and 355%. A developer whose workload is weighted toward generated output can therefore pay more than four and a half times the old output rate at the busiest times of the working day.

Peak pricing turns time into an engineering variable

DeepSeek defines peak hours as 01:00 to 04:00 UTC and 06:00 to 10:00 UTC, Monday through Friday. All other periods are off-peak. The schedule creates an obvious incentive to move batch jobs, large evaluations, synthetic-data generation, and other delay-tolerant work outside those windows. Teams that can shift demand may contain part of the increase; interactive products serving users in Europe and Asia have far less freedom.

This is where the new rate design becomes more than a pricing-page update. Application teams now need time-aware budgets, cost forecasts that reflect their actual traffic profile, and alerts that distinguish a genuine token spike from the same workload crossing into a more expensive window. A flat monthly average can hide a material cost jump if most production calls arrive during the two weekday peak blocks.

The off-peak label deserves scrutiny

Calling the lower tier a discount risks anchoring customers to the new peak price rather than the rate they paid before the change. For V4 Pro output, the so-called off-peak price is not a saving against the former card: it is roughly 128% higher. Cache-miss input rises by about 52% off-peak, while cache-hit traffic sees an even larger multiple. The lower tier is cheaper only relative to DeepSeek's new peak tier.

DeepSeek is entitled to price its infrastructure as it chooses, and greater demand can require capacity investment. Customers are equally entitled to call the change what it is: a broad API price increase wrapped in a demand-management schedule. The model may remain competitive, but the old assumption that DeepSeek is automatically the low-cost answer no longer belongs in a production budget without fresh measurements.

What developers should do now

First, recalculate costs from real usage rather than benchmark prompts. Separate cache hits, cache misses, and output, then map each request timestamp into DeepSeek's peak schedule. Second, measure whether longer reasoning responses are increasing output volume, because output has become the most visible cost driver. Third, test whether repeated prompt prefixes are actually earning cache-hit rates instead of assuming the cache will rescue a large context window.

Teams should also keep model routing and application behavior decoupled. A workload that needs V4 Pro's particular strengths can stay on the model, while routine classification, extraction, or background generation may be moved to a cheaper option or an off-peak queue. The objective is not to abandon DeepSeek on principle; it is to prevent a once-cheap default from becoming an unexamined permanent expense.

Success has a price, and customers should notice it

DeepSeek's pricing move is also a sign of market maturity. The company proved that aggressive inference pricing could win attention and usage. Once that demand existed, it gained room to segment capacity and capture more revenue from time-sensitive customers. That is a familiar commercial arc, even if it sits awkwardly beside the affordable-AI story that helped establish the product.

The right response from developers is neither outrage nor blind acceptance. It is scrutiny. Read the live rate card, calculate the percentage change for your own token mix, make peak exposure visible, and keep alternatives tested. DeepSeek V4 Pro may still earn its place on quality and capability, but after this increase it needs to win that decision on current value, not on a reputation built under yesterday's prices.