DeepSeek launches open-source agent framework and updates flagship model

Chinese AI lab DeepSeek has expanded its offerings by launching DeepSeek Harness v0.1, a new open-source agent framework. This tool provides developers with an alternative to integrated coding agent environments like Anthropic's Claude Code.

Concurrently, DeepSeek has released the official version of its DeepSeek-V4-Pro model. This updated flagship model is heavily focused on agentic workloads and is now available across DeepSeek's web interface, mobile app, and API. It offers native support for the OpenAI Responses API and integrates with Codex.

The release of DeepSeek Harness signifies a move beyond competing solely on model intelligence and token prices, entering the layer that dictates how AI agents use tools, manipulate files, and execute workflows. This competes directly with established developer products such as Anthropic's Claude Code and OpenAI's Codex.

DeepSeek Harness is licensed under the MIT license, with code available for download on GitHub. The framework is designed with a modular premise, allowing virtually any part of the agent runtime to be swapped out as a plugin. While Harness is currently in a developer preview stage with expected breaking changes, its architecture points to a strategy where DeepSeek offers developers not only models but also an open framework for assembling the surrounding systems.

In parallel, DeepSeek has announced an increase in its API pricing. The new pricing model, which includes peak and off-peak rates, will take effect on August 16. Even the discounted off-peak prices are expected to be substantially higher than current rates.