DeepSeek has dramatically changed the economics of its V4 API.
The Chinese AI company has introduced a new peak and off-peak pricing system for its V4 Pro and V4 Flash models, with some API rates increasing by more than 1,000%.
The biggest headline number is a 1,114% increase for V4 Pro cached input tokens during peak hours.
That sounds extraordinary — and it is.
But the headline needs some context.
The largest percentage increase applies specifically to cached input pricing, which was unusually low under the previous pricing structure. Other V4 Pro prices have increased by much smaller percentages, although developers using large amounts of output tokens will still see a substantial jump in costs.
The change could have a noticeable impact on developers building AI agents, coding tools, automation systems and applications that rely heavily on DeepSeek’s API.
What Changed in DeepSeek V4 API Pricing?
DeepSeek has moved away from a single pricing rate and introduced peak and off-peak pricing.
Under the new system, the amount developers pay depends on:
- Which V4 model they use
- Whether input tokens are cached
- Whether input tokens are cache misses
- How many output tokens are generated
- Whether the request happens during peak hours
This creates a more complicated pricing structure than DeepSeek users were previously accustomed to.
The company has effectively separated API usage into different cost levels rather than charging one straightforward rate throughout the day.
The 1,114% Increase Explained
The largest increase applies to DeepSeek V4 Pro cache-hit input tokens during peak hours.
The previous price was approximately:
$0.003625 per million tokens
The new peak price is:
$0.044 per million tokens
That represents an increase of roughly 1,114%.
It is an enormous percentage increase.
However, the absolute price is still only a few cents per million cached input tokens.
That distinction matters.
A percentage increase can look dramatic while the actual dollar impact depends heavily on how many tokens a developer processes.
Why the Percentage Looks So Extreme
The previous cache-hit price was exceptionally low.
That creates a very small starting point.
When the new rate is compared with that unusually low number, the percentage increase becomes enormous.
For example, moving from $0.003625 to $0.044 sounds relatively small in absolute dollar terms.
But mathematically, it represents a dramatic percentage increase.
This is why developers should look at both:
percentage change
and
actual cost per million tokens.
The second figure is usually more useful when deciding whether an API remains affordable.
DeepSeek V4 Pro New Pricing
The V4 Pro changes are substantial across the different token categories.
| V4 Pro Usage | Previous Price | New Off-Peak | New Peak |
|---|---|---|---|
| Cache-hit input | $0.003625/M | $0.022/M | $0.044/M |
| Cache-miss input | $0.435/M | $0.66/M | $1.32/M |
| Output | $0.87/M | $1.98/M | $3.96/M |
The table makes one thing clear.
The cache-hit input increase gets the biggest percentage headline, but output pricing may have a much larger practical effect for applications that generate large amounts of text.
V4 Pro Output Prices Are Also Rising Sharply
DeepSeek V4 Pro output pricing is moving from:
$0.87 per million tokens
to:
$1.98 during off-peak hours
and:
$3.96 during peak hours.
That represents a substantial increase.
For developers running AI agents, coding assistants or reasoning-heavy applications, output tokens can account for a significant portion of total usage.
That means the new pricing structure could materially increase operating costs even when cache-hit input usage is high.
DeepSeek V4 Flash Is Also Getting More Expensive
The price changes are not limited to V4 Pro.
DeepSeek V4 Flash is also receiving higher rates.
Its new pricing is:
| V4 Flash Usage | Previous Price | New Off-Peak | New Peak |
|---|---|---|---|
| Cache-hit input | $0.0028/M | $0.007/M | $0.014/M |
| Cache-miss input | $0.14/M | $0.22/M | $0.44/M |
| Output | $0.28/M | $0.66/M | $1.32/M |
The increases are smaller than the V4 Pro cache-hit jump, but developers using Flash for high-volume workloads will still notice the difference.
What Is Peak and Off-Peak Pricing?
The new system divides API usage into different time periods.
During peak hours, developers pay the higher rate.
During off-peak hours, the price is lower.
For V4 Pro, the peak rate is approximately double the new off-peak rate across the main token categories.
That means developers with flexible workloads can potentially reduce their bills by scheduling heavy jobs during cheaper periods.
This creates a new optimization strategy that did not matter as much under the previous pricing model.
Why Would DeepSeek Introduce Peak Pricing?
DeepSeek has not publicly attributed every aspect of the pricing change to one specific reason.
However, the move comes as demand for advanced AI models continues to grow.
AI inference is expensive.
The more users send requests to a model, the more computing infrastructure is required to process those requests.
Peak pricing can help companies manage demand by encouraging users to move non-urgent workloads into quieter periods.
It is a familiar concept in cloud computing and other infrastructure-heavy industries.
Developers May Need to Think Like Cloud Engineers
The new pricing system changes how developers should approach DeepSeek.
Previously, many users could simply estimate:
tokens × price = cost.
Now they also need to consider:
when the request happens.
For applications running continuously, that can become important.
A developer running a batch-processing pipeline at peak hours could potentially pay significantly more than another developer who schedules the same workload during off-peak periods.
That makes scheduling part of AI cost optimization.
AI Agents Could Feel the Change Most
The pricing increase could be particularly important for AI agents.
Agentic applications often make multiple model calls to complete one task.
A single user request might trigger:
- Planning
- Tool selection
- Search
- Code generation
- Error analysis
- Additional reasoning
- Final response generation
That can produce far more token usage than a normal chatbot conversation.
If output costs rise substantially, agent developers could see their operating expenses increase quickly.
Coding Agents Are Another Concern
AI coding assistants can also generate large amounts of output.
A complex programming task may involve:
- Reading large files
- Understanding project context
- Generating code
- Reviewing code
- Running tests
- Fixing errors
- Repeating the process
Each cycle can consume additional tokens.
Developers using DeepSeek V4 Pro as the reasoning engine behind these systems therefore need to pay close attention to the new output rates.
Cached Tokens Still Matter
DeepSeek’s caching system remains important.
When the same input context can be reused, cache-hit pricing is dramatically lower than cache-miss pricing.
That can be especially useful for applications with long conversations or repeated context.
For example, an AI coding agent may repeatedly send information about the same project.
If that context is cached, the cost structure can be much more favorable than repeatedly processing everything as new input.
The new pricing does not eliminate that advantage.
It simply makes cached input more expensive than before.
The 1 Million Token Context Window Remains
DeepSeek V4 Pro and V4 Flash support a 1 million-token context window.
That is one of the major technical characteristics of the V4 generation.
A large context window allows developers to provide very long inputs, including extensive documents, codebases and conversation history.
For AI agents, this can be particularly useful.
However, a large context window does not mean developers should send unnecessary information.
The new pricing makes efficient context management even more valuable.
DeepSeek V4 Pro Still Supports Advanced Features
The V4 Pro API supports features including:
- Tool calls
- JSON output
- Thinking and non-thinking modes
- Large context windows
- OpenAI-compatible API access
- Anthropic-compatible API access
That makes it suitable for a broad range of AI applications.
The price increase therefore does not represent a reduction in the model’s capabilities.
Instead, it changes the economics of accessing those capabilities.
Is DeepSeek Still Cheap?
This is where the situation becomes interesting.
Even after the increase, DeepSeek’s API pricing remains competitive with many advanced AI models.
For example, the new V4 Pro peak output price of $3.96 per million tokens is still below the price of some competing frontier models.
That means the statement that DeepSeek is suddenly “expensive” needs qualification.
It is more accurate to say that DeepSeek has become significantly more expensive relative to its previous pricing advantage.
DeepSeek’s Biggest Advantage Was Its Price
DeepSeek gained enormous attention partly because of its ability to deliver strong model performance at unusually low prices.
That pricing strategy made it attractive to developers who wanted to run large workloads without paying premium API rates.
The new structure changes that equation.
DeepSeek can still be inexpensive compared with some competitors.
But the difference is no longer as extreme as it was.
For price-sensitive developers, that matters.
The New Pricing Could Change Model Selection
Developers now have more reasons to choose between V4 Pro and V4 Flash based on workload requirements.
A simple task may not need the more expensive Pro model.
A high-volume application might benefit from Flash.
A complex reasoning task may justify the additional cost of Pro.
That creates a more conventional model-selection strategy:
Use the cheapest model that reliably completes the task.
Developers May Move Heavy Jobs to Off-Peak Hours
The new pricing creates a straightforward cost-saving opportunity.
If a workload does not need immediate results, developers can schedule it during off-peak hours.
Examples could include:
- Batch document processing
- Dataset analysis
- Large-scale summarization
- Offline code analysis
- Content classification
- Testing
- Evaluation workloads
Real-time applications obviously have less flexibility.
But businesses running scheduled workloads may be able to reduce the impact of the price increase through better timing.
What About Real-Time Applications?
Real-time applications face a different problem.
A chatbot cannot necessarily tell a user to come back later because API pricing is cheaper.
The same applies to:
- Customer-support agents
- Coding assistants
- Search systems
- Interactive applications
- Voice assistants
- Real-time automation
For these products, peak-hour pricing may become a direct operating expense.
Developers may therefore need to optimize prompts, reduce unnecessary output and use smaller models where possible.
Prompt Efficiency Matters More Now
The pricing change could encourage developers to become more careful about token consumption.
Instead of sending enormous prompts on every request, applications can:
- Remove redundant context
- Cache repeated information
- Summarize old conversations
- Reduce unnecessary output
- Use structured responses
- Route simple tasks to cheaper models
- Schedule non-urgent workloads off-peak
These practices were already useful.
The new pricing simply makes them more financially important.
Will Developers Leave DeepSeek?
Some users may.
Price-sensitive developers who selected DeepSeek primarily because it was extremely cheap now have a reason to reconsider their choices.
Others may stay because the model’s quality, context window and agent capabilities continue to justify the cost.
The answer will depend heavily on the workload.
A developer processing millions of tokens every day will care much more about the price increase than someone using the API occasionally.
The Competitive Landscape Is Changing
DeepSeek’s pricing move arrives during a period of intense competition among AI model providers.
Companies are competing on:
- Intelligence
- Speed
- Context length
- Tool use
- Coding ability
- Reliability
- API price
- Infrastructure
- Agent capabilities
Price has become one of the easiest ways for developers to compare models.
DeepSeek’s new rates could therefore influence how companies position their own APIs.
Could Competitors Benefit?
Potentially.
If DeepSeek’s biggest advantage was its low cost, competing providers now have an opportunity to narrow that gap.
Developers may compare DeepSeek with alternatives based on both price and performance.
A model that is slightly more expensive but significantly better at a specific workload could become more attractive.
That is especially true for businesses that care about total cost per successful task rather than cost per token alone.
The Price Increase Does Not Mean DeepSeek Is Failing
It would be a mistake to interpret the increase as evidence that DeepSeek is abandoning its competitive strategy.
The company continues to develop large models, agent capabilities and infrastructure.
The pricing change may instead reflect a shift from aggressive low-cost positioning toward a more sustainable commercial model.
That could be a normal step for a fast-growing AI company.
What Developers Should Do Now
Developers using DeepSeek V4 APIs should review their usage rather than immediately abandon the platform.
Start by checking:
- Which model is being used
- Average input tokens
- Average output tokens
- Cache-hit percentage
- Cache-miss percentage
- Peak-hour usage
- Off-peak usage
- Cost per completed task
This gives a much clearer picture than simply looking at the headline 1,114% increase.
Calculate Cost Per Task, Not Just Cost Per Token
Token prices are useful, but they are not the whole story.
Suppose one model costs twice as much per million tokens but completes a task using half as many tokens.
The actual cost may be similar.
Likewise, a cheaper model may require multiple retries or additional verification.
That can make the supposedly cheaper model more expensive in practice.
Developers should therefore compare cost per successful task.
The Bigger Picture
DeepSeek’s latest pricing decision marks an important moment in the AI API market.
For years, the company helped push the industry toward lower inference costs.
Now its V4 models are moving toward a more differentiated pricing structure based on model capability, caching and time of use.
The 1,114% figure is certainly eye-catching.
But it does not mean every DeepSeek API request is suddenly 11 times more expensive.
The real impact depends on what model is being used, how many tokens are processed, whether those tokens are cached and whether the request happens during peak hours.
Read More:- OpenAI Is Quietly Preparing for History’s Biggest Tech IPO | Affitronix
Bottom Line
DeepSeek V4 API prices have jumped sharply, with some rates increasing by up to 1,114%.
The biggest increase affects V4 Pro cached input tokens during peak hours, rising from about $0.003625 to $0.044 per million tokens.
But developers should not interpret the headline as an across-the-board 1,114% increase.
V4 Pro input, output and V4 Flash pricing have all changed by different amounts, while the new peak/off-peak system adds another variable to the calculation.
For developers, the biggest practical changes are likely to be higher output costs, more expensive cache hits and the need to consider when workloads are executed.
DeepSeek remains competitively priced compared with many frontier AI APIs.
But its enormous cost advantage has narrowed.
For companies building AI agents, coding tools and high-volume applications, the new pricing means one thing above all:
AI cost optimization just became much more important.
FAQ
Why did DeepSeek increase its API prices?
DeepSeek has introduced a new peak/off-peak pricing structure for its V4 models. The company has not attributed every aspect of the change to one specific reason, but the new structure changes how API usage is priced based on model, token type and time of use.
What is the 1,114% DeepSeek price increase?
The 1,114% figure refers to the increase in DeepSeek V4 Pro cache-hit input pricing during peak hours, which moved from approximately $0.003625 to $0.044 per million tokens.
How much does DeepSeek V4 Pro cost now?
Under the new structure, V4 Pro costs approximately $0.022 per million cache-hit input tokens off-peak and $0.044 during peak hours. Cache-miss input costs $0.66 off-peak and $1.32 at peak, while output costs $1.98 off-peak and $3.96 at peak.
How much did DeepSeek V4 Pro output pricing increase?
V4 Pro output pricing increased from $0.87 per million tokens to $1.98 during off-peak hours and $3.96 during peak hours.
Did DeepSeek V4 Flash prices also increase?
Yes. V4 Flash also received higher rates. Its new peak output price is $1.32 per million tokens, compared with the previous $0.28 rate.
What is DeepSeek’s peak/off-peak pricing?
DeepSeek now charges different rates depending on when API requests are processed. Peak periods have higher prices, while off-peak periods are cheaper. This allows developers with flexible workloads to potentially reduce costs by scheduling requests outside peak periods.
Is DeepSeek still cheap after the price increase?
DeepSeek remains competitive with many advanced AI APIs, even after the increase. However, its price advantage is significantly smaller than it was under its previous ultra-low pricing structure.
Why are cached tokens important?
Cached tokens are input tokens that can be reused from previous context. They are generally much cheaper than cache-miss tokens, making caching particularly useful for applications with repeated prompts, long conversations and AI agents.
Will the price increase affect AI agents?
Yes, potentially. AI agents often make multiple model calls and generate large amounts of output. Higher output and peak-hour prices can therefore increase the cost of agentic workflows significantly.
Should developers stop using DeepSeek V4?
Not necessarily. Developers should compare the new cost against model quality and the cost per successful task. For some applications, V4 Pro may remain attractive, while other workloads may be better served by V4 Flash or another provider.
