DeepSeek has officially released DeepSeek V4 Pro 0813, the latest version of its flagship AI model and one of the most important open-weight model releases of August 2026.
The new release focuses heavily on coding, reasoning, AI agents, and long-context workloads. It keeps the V4 architecture's massive one-million-token context window while delivering significant improvements over the earlier V4 Pro preview.
However, there is one major change users should know about: DeepSeek has substantially increased API pricing for V4 Pro.
What Is DeepSeek V4 Pro?
DeepSeek V4 Pro is DeepSeek's flagship Mixture-of-Experts AI model. The model uses approximately 1.6 trillion total parameters, with around 49 billion active parameters, and supports a context window of up to one million tokens.
The model is available through DeepSeek's website, mobile application, and API. The latest general-availability version is identified as DeepSeek-V4-Pro-0813.
DeepSeek also publishes the model weights under the MIT license, making V4 Pro particularly interesting for developers and organizations that want more control over how an AI model is deployed.
What's New in DeepSeek V4 Pro 0813?
The August release is more than a simple model refresh. DeepSeek has improved the model's performance on agentic and coding workloads while making it more token-efficient.
Stronger AI Agent Capabilities
One of the biggest improvements is agentic performance.
DeepSeek V4 Pro 0813 is designed to perform multi-step tasks involving tools, planning, code execution, and longer workflows. Independent testing from Artificial Analysis found that most of the improvement over the April version came from stronger agentic capabilities.
The model also supports the Responses API and integrations designed for coding-agent workflows.
1 Million Token Context Window
DeepSeek V4 Pro retains a one-million-token context window.
This is one of the model's biggest advantages for developers working with large repositories, long documents, extensive conversations, or large collections of source material.
A one-million-token context allows an AI system to process far more information in a single request than many traditional models.
Improved Coding
Coding remains one of DeepSeek V4 Pro's strongest use cases.
The model is designed to work with large codebases and complex software-engineering tasks. Its long context window also makes it suitable for analyzing large repositories without repeatedly splitting the project into smaller pieces.
Artificial Analysis reports that the new V4 Pro 0813 reached a score of 53 on its Intelligence Index, eight points higher than the April V4 Pro preview.
Better Token Efficiency
Another interesting improvement is efficiency.
Artificial Analysis found that V4 Pro 0813 used approximately 30% fewer output tokens than the earlier V4 Pro version while running its Intelligence Index evaluation.
That matters because fewer generated tokens can reduce both latency and operating costs for applications that use AI agents extensively.
DeepSeek V4 Pro Technical Specifications
| Feature | DeepSeek V4 Pro 0813 |
|---|---|
| Architecture | Mixture of Experts |
| Total parameters | Approximately 1.6 trillion |
| Active parameters | Approximately 49 billion |
| Context window | 1 million tokens |
| Weights | Open weights |
| License | MIT |
| Main strengths | Coding, reasoning, agents and long-context tasks |
| Availability | Web, app and API |
DeepSeek's published V4 architecture also uses a hybrid attention design intended to reduce the computational and memory requirements of very long contexts.
DeepSeek V4 Pro Pricing
Pricing is one of the biggest changes with the new release.
Before the latest price change, V4 Pro was available at approximately $0.435 per million input tokens and $0.87 per million output tokens under the earlier pricing structure.
Beginning in August 2026, DeepSeek introduced a new peak and off-peak pricing structure.
| V4 Pro API | Off-Peak | Peak |
|---|---|---|
| Input, cache miss | $0.66 / 1M | $1.32 / 1M |
| Input, cache hit | $0.022 / 1M | $0.044 / 1M |
| Output | $1.98 / 1M | $3.96 / 1M |
The new peak pricing represents a major increase compared with the earlier promotional rates. Reuters reported that the new V4 Pro pricing is substantially higher than V4 Flash, although it remains competitive with many frontier models.
Developers should therefore check DeepSeek's current API pricing before estimating the cost of a production application.
Is DeepSeek V4 Pro Still Cheap?
Yes, relative to many frontier AI models, but the gap has narrowed considerably.
The previous V4 Pro pricing was one of DeepSeek's biggest advantages. The new pricing is significantly higher, especially during peak hours.
Even after the increase, DeepSeek V4 Pro can remain attractive for developers who need strong coding and agent capabilities at a lower price than some premium proprietary models.
DeepSeek V4 Pro vs DeepSeek V4 Flash
| Feature | V4 Pro | V4 Flash |
|---|---|---|
| Positioning | Flagship | Lower-cost model |
| Total parameters | ~1.6T | ~284B |
| Active parameters | ~49B | ~13B |
| Context | 1M tokens | 1M tokens |
| Best for | Complex reasoning and agents | High-volume and cost-sensitive tasks |
Artificial Analysis found that V4 Pro 0813 scored only one point higher than V4 Flash 0731 on its Intelligence Index, while both models achieved 79% on Terminal-Bench 2.1 in the cited evaluation.
This makes V4 Flash particularly interesting for developers who care more about cost and throughput than maximum model capability.
DeepSeek V4 Pro vs Claude
DeepSeek is increasingly targeting the same coding and agent workloads as leading Claude models.
In DeepSeek's published comparisons, Claude Fable 5 was ahead by an average of around 5.3% across nine agent benchmarks where both models were scored. However, the price difference was enormous.
Reported API pricing puts Claude Fable 5 at around $10 per million input tokens and $50 per million output tokens, compared with $0.435 and $0.87 for the earlier V4 Pro pricing.
The exact comparison depends on the model version and pricing period, but the broader point remains: DeepSeek is competing aggressively on the intelligence-per-dollar metric.
DeepSeek V4 Pro vs GPT-5.6
DeepSeek V4 Pro is also competing with OpenAI's latest models, particularly in coding and agentic software development.
Artificial Analysis currently places GPT-5.6 Sol higher on its Intelligence Index, with a score of 61 compared with 53 for V4 Pro 0813.
However, V4 Pro has a major advantage in openness and deployment flexibility. Its weights are available under the MIT license, while OpenAI's frontier models are proprietary.
| Feature | DeepSeek V4 Pro | GPT-5.6 |
|---|---|---|
| Open weights | Yes | No |
| Context | 1M tokens | Varies by model/product |
| Coding | Very strong | Frontier-level |
| Agent workflows | Strong | Strong |
| Self-hosting flexibility | High | Limited |
| Best advantage | Open weights and cost efficiency | Overall frontier intelligence |
DeepSeek V4 Pro vs Other AI Models
DeepSeek V4 Pro occupies an unusual position in the current AI market.
It is not simply trying to become the most intelligent model available. Instead, it combines high-end reasoning and coding capabilities with open weights, a very large context window, and relatively low inference costs.
This makes it especially attractive for developers building their own AI products.
What Is DeepSeek V4 Pro Best For?
- Large codebases: The one-million-token context is useful for analyzing large repositories.
- AI coding agents: The model is designed for multi-step software-engineering workflows.
- Long documents: Large context allows extensive source material to be processed together.
- Reasoning: V4 Pro supports extended reasoning workflows.
- AI applications: Open weights provide more flexibility for developers.
- Cost-sensitive workloads: Despite the recent price increase, the model remains competitive with many frontier alternatives.
Who Should Use DeepSeek V4 Pro?
DeepSeek V4 Pro is primarily aimed at developers, researchers, AI engineers, and businesses building applications around large language models.
It is particularly interesting if you need to process very large contexts or want more control over your model deployment.
For casual users who simply want an AI chatbot, the DeepSeek web and mobile applications may be more convenient than working directly with the API.
Is DeepSeek V4 Pro Open Source?
DeepSeek describes the V4 models as open-weight models, and the V4 Pro weights are published under the MIT license.
However, it is useful to distinguish open weights from the broader term "open source." Having model weights available does not necessarily mean that every component of the training process, training data, infrastructure, and development process is publicly reproducible.
For developers, the important advantage is that the weights can be downloaded and used under the published license terms.
What Are the Limitations?
DeepSeek V4 Pro is powerful, but it is not the perfect model for every workload.
Its enormous model size makes local deployment challenging. Although only a fraction of the total parameters are active for each token, running the full model still requires substantial hardware resources.
The model is also primarily attractive for text-based reasoning and coding workloads. Developers looking for a single model with the broadest possible native multimodal capabilities may prefer alternatives designed specifically around image, audio, and video understanding.
Finally, benchmark performance varies significantly by task. A model that performs extremely well on coding agents may not necessarily be the best choice for writing, visual reasoning, or everyday conversations.
DeepSeek V4 Pro: Pros and Cons
Pros
- One-million-token context window.
- Open weights under the MIT license.
- Strong coding and agentic performance.
- Very large total model capacity.
- Improved token efficiency.
- Competitive API economics compared with some frontier models.
- Available through web, app and API.
Cons
- API prices increased significantly in August 2026.
- The full model is extremely large for local deployment.
- Not necessarily the best model for every benchmark or use case.
- Independent evaluation is still important when choosing it for production.
Is DeepSeek V4 Pro Worth It?
For developers, the answer can be yes.
The combination of a one-million-token context window, strong coding capabilities, agentic improvements, and open weights makes V4 Pro one of the most interesting AI models available in 2026.
The recent price increase does reduce one of DeepSeek's biggest historical advantages, but the model remains significantly cheaper than some premium frontier alternatives.
The more important question is therefore not simply whether V4 Pro is cheap. It is whether the model provides enough performance for your particular workload at an acceptable cost.
Final Verdict
DeepSeek V4 Pro 0813 is a major upgrade for developers looking for an open-weight model capable of handling serious coding and agentic workloads.
Its one-million-token context window remains one of its strongest features, while the latest release improves agent performance and token efficiency.
The biggest surprise is pricing. DeepSeek has moved away from the extremely low promotional rates associated with the earlier V4 Pro release and introduced substantially higher peak-hour prices.
Even so, V4 Pro remains an important competitor because it combines strong capabilities with open weights and a large context window.
For developers who prioritize open models, long context, coding, and AI agents, DeepSeek V4 Pro deserves serious consideration.
For users who prioritize maximum overall intelligence, the latest proprietary frontier models may still have an advantage.
The most interesting part of DeepSeek V4 Pro is therefore not that it beats every competing model. It is that it continues to narrow the gap while giving developers far more freedom over how the model is used.
Frequently Asked Questions
What is DeepSeek V4 Pro?
DeepSeek V4 Pro is DeepSeek's flagship open-weight Mixture-of-Experts model, designed for reasoning, coding, long-context processing, and AI-agent workflows.
What is DeepSeek V4 Pro 0813?
V4 Pro 0813 is the general-availability version of DeepSeek V4 Pro released in August 2026.
How much does DeepSeek V4 Pro cost?
Pricing changed significantly in August 2026. Current peak API pricing for V4 Pro is $1.32 per million input tokens and $3.96 per million output tokens, with lower off-peak rates.
Does DeepSeek V4 Pro have a 1M context window?
Yes. DeepSeek V4 Pro supports a context window of approximately one million tokens.
Is DeepSeek V4 Pro open source?
DeepSeek publishes V4 Pro's weights under the MIT license. The more precise description is an open-weight model rather than assuming that every part of its training process is open.
Is DeepSeek V4 Pro good for coding?
Yes. Coding and software-engineering workflows are among its strongest use cases, particularly when combined with its large context window and agent capabilities.
