The model, named DeepSeek-V4-Pro-0813, is now available through DeepSeek's website, mobile application, and API, according to the company.
The launch follows the release of the preview version of V4 Pro and represents a further upgrade in the model's ability to handle complex tasks. DeepSeek has also introduced support for the Responses API and Codex integration, providing developers with enhanced tools for building AI agent applications.
Unlike traditional AI models that mainly respond to user queries, AI agents are capable of breaking down complex tasks, using external tools, executing operations, checking results and making decisions about subsequent steps. Such capabilities enable AI systems to take on more complex tasks with less human intervention.
According to benchmark results released by DeepSeek, the new model has made substantial progress in software engineering and other agent-related tasks.
On the DeepSWE benchmark, which evaluates an AI model's ability to resolve real-world software engineering problems, DeepSeek-V4-Pro-0813 scored 62.7, compared with 12.8 for the preview version.
The model also demonstrated strong performance in terminal operations and cybersecurity-related tasks, according to the company.
In addition to improving performance, DeepSeek has maintained relatively low API prices for the new model. The current price is 3 yuan (about $0.42) per million input tokens and 6 yuan per million output tokens.
DeepSeek said it will adjust its API pricing from August 17, introducing different rates for peak and off-peak periods. During peak hours on weekdays, from 9 a.m. to noon and from 2 p.m. to 6 p.m. Beijing time, the price for output tokens will rise to 27 yuan per million, while the off-peak price will be half that amount.
The launch comes as AI applications are developing from question-and-answer services toward more complex tasks involving coding, tool use, and task execution. AI agents are expected to play a growing role in areas such as software development and other professional applications.
Earlier, it was reported that Google DeepMind introduced a new sign-language-to-text model designed to help Deaf and hard-of-hearing users communicate through their phones.