DeepSeek has unveiled Harness v0.1, an open-source development environment for AI agents, alongside the final version of its V4 Pro model.

https://twitter.com/deepseek_ai/status/2087887408440164663

Designed as an open alternative to Claude Code, Harness is distributed under the MIT license and is currently available in a preliminary iteration for developers on GitHub. This environment operates as a layer over large language models (LLMs), equipping agents with tools to interact with codebases, terminals, and external tools.

Simultaneously, the company has rolled out the final version of DeepSeek V4 Pro, enhancing its agent capabilities and adding support for the OpenAI Responses API.

An Open Competitor to Claude Code

Harness is intended for scenarios where the model not only generates code on demand but also autonomously executes a series of actions within a working environment.

DeepSeek markets it as a minimalist agent framework that can be utilized with custom models and tailored for specific tasks.

According to the company, Harness was employed for testing the agent capabilities of V4 Pro across several public benchmarks.

DeepSeek is betting not only on the framework itself but also on the combination of "model + agent shell": Harness is expected to assist V4 Pro in utilizing tools more effectively and executing longer sequences of actions.

New Agent Features in V4 Pro

The final version of V4 Pro is available through applications, web versions, and API. DeepSeek reports a significant increase in performance for agent tasks. The model now includes native support for the Responses API and integration with Codex, with reasoning levels selectable from three modes: low, high, and max.

In tests published by the company, V4 Pro achieved scores of 87.9 in Terminal Bench 2.1, 62.7 in DeepSWE, and 74.1 in Toolathlon-Verified.

Source: DeepSeek.

API Pricing to Increase

With the full release of DeepSeek, the company is revising its pricing structure. A dynamic model will be introduced, featuring peak and off-peak hours with rates reduced by half during non-peak times.

— red šŸ“Œ (@redkendl) August 13, 2026

These new pricing tiers will take effect on August 16.

It is worth noting that the formation of the DeepSeek team for the development of Harness was announced in May.