DeepSeek Unveils its Flagship AI, V4 Pro, with a 1M Token Context Window
Chinese AI leader, DeepSeek, has announced the general availability of its premier model, DeepSeek V4 Pro (version 0813). The model, which was in preview for nearly four months, was released on August 13, 2026. It is now accessible across the company’s app, web interface, and API. This launch marks a significant milestone in DeepSeek’s campaign to regain its position against rapidly advancing domestic competitors.
The V4 Pro’s distinguishing feature is its emphasis on AI agent capabilities. These autonomous, multi-step task execution abilities require no human intervention and are designed for complex workflows in software development, coding, and automation. The model is constructed on a Mixture-of-Experts architecture with 1.6 trillion total parameters and 49 billion active parameters per token. It supports a context window of up to 1 million tokens and can generate outputs as long as 384,000 tokens in a single request, with selectable thinking effort levels (low, high, and max).
Concurrent with the model’s release, DeepSeek launched DeepSeek Harness v0.1, an open-source agent framework on GitHub under the MIT license. The company also revised its pricing structure, introducing dynamic peak and off-peak API billing from August 16, with off-peak rates at 50% of peak-hour pricing. Benchmark results reveal the model scoring 87.9 on Terminal Bench 2.1 and 62.7 on DeepSWE, among other agent-focused tests. The launch comes amidst rumors that DeepSeek is also developing its own AI chip to decrease dependence on Nvidia and Huawei for compute.
Source: Quartz — DeepSeek officially launches V4-Pro AI model in August 2026