DeepSeek launches V4 Pro model and Harness agent framework

DeepSeek has launched its flagship V4 Pro AI model alongside DeepSeek Harness, an open-source execution framework designed to facilitate the deployment of AI agents. The V4 Pro model features a one-million-token context window and improved capabilities for complex, multi-step tasks, such as coding and tool use. To support these advancements, the company introduced DeepSeek Harness, a modular platform where components like models, tools, and sandboxes function as swappable plugins. This framework aims to bridge the gap between model reasoning and reliable engineering execution in real-world environments.

Accompanying these technical releases is a significant restructuring of DeepSeek’s API pricing. Starting in mid-August, the company is implementing a peak and off-peak pricing model to better manage resource distribution. Consequently, API costs for certain models will increase substantially, with some rates rising by up to 1,100 percent during peak hours. Despite these hikes, DeepSeek maintains that its pricing remains competitive compared to major industry rivals. The V4 Pro model has also been integrated into China’s national supercomputing network to accelerate AI agent development in scientific research and industrial sectors.