DeepSeek challenges the leaders: release of the Harness agent framework and the final version of V4 Pro
Chinese AI giant DeepSeek has made a double strategic move by unveiling two significant products to the public simultaneously. These are the open-source framework for developing AI agents, Harness v0.1, and the long-awaited final version of the V4 Pro model. This step clearly demonstrates the company's intention to strengthen its position not just as a provider of language models, but also as a creator of a full-fledged ecosystem for autonomous agents.
Harness: An Open Alternative to Proprietary Solutions
The Harness framework, distributed under the MIT license, is positioned as a direct competitor to Anthropic's popular tool, Claude Code. Unlike closed solutions, Harness offers developers a flexible layer on top of large language models (LLMs), providing the agent with tools to interact with the codebase, terminal, and external services. It is not just a code generator, but an environment for executing complex, multi-step tasks in a working environment. Notably, it was Harness that was used for internal testing of V4 Pro's agentic capabilities on public benchmarks, highlighting the synergy between these products.
V4 Pro: A New Level of Agentic Capability and Performance
The final version of V4 Pro marks a significant leap in agentic capabilities. The model has native support for the OpenAI Responses API and integration with Codex, simplifying its adoption into existing development pipelines. Flexible configuration of reasoning levels (low, high, max) allows developers to balance between speed and depth of analysis. Benchmark results are impressive: 87.9 points in Terminal Bench 2.1, 62.7 in DeepSWE, and 74.1 in Toolathlon-Verified. These figures confirm that V4 Pro can compete with the best models in tasks involving autonomous code execution and tool use.
API Economics: Dynamic Pricing
From a commercial standpoint, the company is transitioning to a new pricing model for its API. Dynamic pricing is being introduced with peak and off-peak hours, with costs during off-peak hours being twice as low. Although actual prices will roughly double, they still remain among the most attractive on the market, making DeepSeek a serious competitor to OpenAI and Anthropic. The new rates will take effect on August 16.
My analysis: DeepSeek's strategy is obvious — they are not just releasing a model, but creating a closed-loop "model + toolkit" ecosystem, which is critical for attracting enterprise developers. The open-source nature of Harness is a strong move that could quickly build a community around the framework and establish standards. In the long term, it is precisely such combinations, rather than individual models, that will determine leadership in the AI industry.