The price of the DeepSeek API is changing shape from August 16, 2026, 16:00 UTC. The Chinese lab, known until now for its unbeatable rates, is introducing two-tier pricing: peak and off-peak hours, with increases ranging from 50% to over 1,000% depending on the model and token type. For an SMB that built an automation or a chatbot on DeepSeek because of its rock-bottom price, September's bill could come as a surprise if nothing gets adjusted.
At a glance
- DeepSeek's new V4 pricing grid took effect on August 16, 2026, 16:00 UTC (source: official DeepSeek documentation,
api-docs.deepseek.com). - V4-Flash output prices go from a flat rate of $0.28 per million tokens to $0.66 off-peak and $1.32 at peak, up to 4.7 times more expensive at peak hours.
- Peak hours are set at 01:00-04:00 and 06:00-10:00 UTC; the rest of the day is billed at the off-peak rate, half as expensive (source: official DeepSeek documentation).
- According to Reuters, the overall increase ranges from 50% to 1,100% depending on the model, the token type, and the time of use.
- Despite the hike, DeepSeek remains far cheaper than its competitors: $1.32 per million output tokens at peak, versus roughly $15 for Kimi K3 (Moonshot AI) and $50 for Claude 5 (Anthropic), according to PYMNTS.
- For an SMB, the right move is to shift non-urgent batch processing to off-peak hours rather than switching providers in a hurry.
What actually changes in the pricing grid
DeepSeek justifies the change by a need to better balance load on its infrastructure. In its official announcement, the company says it wants to "allocate resources more reasonably" through peak/off-peak pricing, "with off-peak prices set at half of the peak-hour prices, encouraging users to schedule their tasks based on actual usage."
Here is how the two main models of the V4 family change, according to DeepSeek's official documentation:
| Model | Category | Old rate (flat) | New off-peak rate | New peak rate |
|---|---|---|---|---|
| V4-Flash | Input (cache miss) | $0.14 | $0.22 | $0.44 |
| V4-Flash | Output | $0.28 | $0.66 | $1.32 |
| V4-Pro | Input (cache miss) | $0.435 | $0.66 | $1.32 |
| V4-Pro | Output | $0.87 | $1.98 | $3.96 |
Prices in dollars per million tokens. Source: official DeepSeek documentation (api-docs.deepseek.com/quick_start/pricing).
The off-peak rate nearly doubles the price compared with the old flat grid. The peak rate multiplies it by 4.7. For a regular, non-urgent workload, staying within off-peak hours sharply limits the impact of this increase.
Why this increase now
DeepSeek never hid the fact that its prices, far below those of American labs, could not stay at the same level forever as demand grows. Differentiated pricing is a way to manage server load without going back to a single flat rate, higher for everyone, all the time.
It is also worth putting this increase in perspective: even at the peak rate, DeepSeek remains highly competitive against the market's leading players.
Comparison reported by PYMNTS. The models compared are not necessarily of strictly identical tier: verify on your own use cases before any budget decision.
In other words, the new grid reshuffles the deck, but does not erase the price gap that made DeepSeek's success with technical teams looking for low-cost AI.
A simple trick: play with time zones
DeepSeek's peak hours are expressed in UTC: 01:00-04:00 and 06:00-10:00. Converted to Paris time (UTC+2 in August), that is roughly 03:00-06:00 and 08:00-12:00. In practice, for an SMB working regular office hours, the morning until noon partly falls within peak hours, but the whole afternoon and evening are billed at the off-peak rate.
Worth checking before automating
This conversion is approximate and depends on the time of year (daylight saving or standard time). Always check the exact UTC time shown by your monitoring tool before scheduling an automated task, rather than relying on mental math.
What this actually changes for an SMB
Audit your current AI usage
Separate urgent from deferrable
Reschedule batch jobs
Don't depend on a single low-cost provider
GDPR reminder
This price increase changes nothing about data compliance. DeepSeek's hosted API remains a Chinese service: still do not send European customer or employee data to it without validation from your data protection officer. We cover this point in detail in our article on DeepSeek V4 and GDPR.
FAQ
Why is DeepSeek raising its prices now?
According to the company, the goal is to better balance load on its infrastructure by encouraging users to schedule non-urgent tasks during off-peak hours, rather than concentrating everything on the same time slots. Growing demand for its V4 models likely also plays a role, even though DeepSeek does not state this explicitly in its documentation.
What is "peak / off-peak" pricing for an AI API?
It is a system where the price per token varies by time of day: more expensive during high-demand periods (peak hours), cheaper the rest of the time (off-peak hours). At DeepSeek, peak hours are set at 01:00-04:00 and 06:00-10:00 UTC, and the off-peak rate is half the peak rate.
Is DeepSeek still cheaper than ChatGPT or Claude after this increase?
Yes, by a wide margin. Even at the peak rate, DeepSeek V4-Flash costs $1.32 per million output tokens, versus roughly $15 for Kimi K3 and $50 for Claude 5, according to figures reported by PYMNTS. The gap narrows, but DeepSeek keeps a clear price advantage.
How can an SMB benefit from off-peak hours without switching tools?
By simply rescheduling when non-urgent tasks run (summaries, data extraction, batch processing) in your automation tool or workflow orchestrator, without changing provider or model. Tasks that require a real-time response, like customer support, cannot be shifted this way.
In conclusion
This increase confirms a reality few SMBs anticipate: a low price on an AI service is never guaranteed to last. Even after this revision, DeepSeek remains one of the most economical options on the market, but the bill for poorly planned usage can now climb quickly. A simple audit of your usage, and shifting non-urgent tasks to off-peak hours, is enough in most cases to limit the impact. To dig deeper into choosing an AI model that fits your budget, read our article on DeepSeek V4 and GDPR or browse all our Mag resources.
