DeepSeek challenges the leaders: open-source Harness framework and final release of V4 Pro
Chinese AI giant DeepSeek has made a double strategic move, presenting the developer community with two interconnected new products: an open environment for creating AI agents, Harness v0.1, and the long-awaited final version of its flagship model, V4 Pro. This is not just a lineup update, but a bid to build its own ecosystem capable of competing with proprietary solutions at the level of Claude Code.
Harness: A New Philosophy for Agent Development
The key announcement is the Harness framework, distributed under the MIT license. Unlike closed solutions, DeepSeek offers a minimalist yet powerful tool that works as an overlay on any LLM. Harness provides the agent with a full range of tools for interacting with the codebase, terminal, and external services. My assessment: this is a bold step that could radically democratize access to agentic development, allowing teams to create their own highly efficient pipelines without being tied to a specific vendor.
V4 Pro: Betting on Autonomy
The final version of V4 Pro is a logical continuation of this strategy. The model has received a significant boost in agentic capabilities, confirmed by impressive benchmark results: 87.9 points in Terminal Bench 2.1, 62.7 in DeepSWE, and 74.1 in Toolathlon-Verified. Of particular note is native support for the OpenAI Responses API and integration with Codex, making the model extremely flexible for developers accustomed to the OpenAI ecosystem. The flexible reasoning level selection system (low, high, max) allows for precise tuning of the balance between speed and depth of analysis.
New Pricing Policy
Alongside the release, DeepSeek announces a revision of API tariffs. Starting August 16, a dynamic pricing model will be introduced with peak and off-peak hours, where the cost during off-peak hours will be twice as low. However, as many observers note, this essentially means a twofold increase in base rates. Nevertheless, even with this taken into account, DeepSeek's offering remains one of the most cost-effective on the market, preserving their competitive advantage.
My conclusion: The Harness + V4 Pro combination is a well-thought-out ecosystem aimed at capturing market share in the autonomous agent segment. The success of this strategy will depend on how actively third-party developers adopt the open framework and whether they can achieve results comparable to closed alternatives.