Editorial illustration for DeepSeek Harness Launches as Open Source Rival to Claude Code
DeepSeek-V4-Pro: Open Source Claude Code Rival
DeepSeek pushed two releases out the door on Thursday, and only one of them is a language model. The Chinese AI lab launched DeepSeek-V4-Pro, its updated flagship focused on agentic work, now live across the web interface, mobile app and API with native support for OpenAI's Responses API and Codex integration. Alongside it came DeepSeek Harness v0.1, an open-source agent harness released under the MIT license, with code already up on GitHub for developers to download and run. It's a direct alternative to the integrated coding-agent setups that Anthropic built around Claude Code, and DeepSeek designed it around a modular structure where nearly every piece of the agent runtime can be swapped out as a plugin.
There's a catch for anyone using V4 through the API, though. DeepSeek is scrapping its flat pricing structure in favor of peak and off-peak rates starting August 16 at 16:00 UTC, and even the cheaper off-peak numbers will run well above what developers pay today. That pricing shift matters less on its own than in combination with Harness, since DeepSeek is now angling for control over how agents actually operate, not just how smart or cheap the underlying model is.
The combination is significant because DeepSeek is no longer competing solely over model intelligence and token prices. With Harness, it is moving into the layer that determines how models use tools, manipulate files, maintain sessions and execute long-running agent workflows — territory where Anthropic’s Claude Code and other coding agents have become increasingly important developer products.
Why this matters
DeepSeek is no longer content selling weights and letting the ecosystem build the tooling around them. By shipping Harness alongside V4-Pro, the lab is copying Anthropic's playbook: control the model and the environment it runs in, then publish benchmarks that reflect both. That's worth pausing on.
When DeepSeek says V4-Pro-0813 was tested in Harness's "minimal mode," it's telling us the agent results aren't clean model-versus-model comparisons, they're model-plus-scaffolding numbers. For developers deciding whether to adopt Harness or stick with Claude Code, that distinction matters more than the headline scores. Open-sourcing the harness is a real gift to teams who want to inspect or modify their agent loop instead of trusting a black box, and higher API pricing on V4-Pro signals DeepSeek thinks its agentic performance now justifies charging like a frontier lab.
Founders building on either stack should read the benchmarks with that caveat in mind, and watch whether independent testers can reproduce these numbers outside DeepSeek's own harness before treating them as gospel.
Related Reading
Browse all Anthropic news →
Open Source D&B rebuilds 642 million‑business database after AI agents hit limits
Open Source OpenAI's GPT Transcribe Cuts Error Rate to 3.31%, Improving on GPT-4o
Open Source Study: 6.7% of Deepfake Requests on Hugging Face Targeted Children
Related Topic Anthropic says Claude writes >90% of its code; AI pause button urged
LLMs & Generative AI
Related Topic Claude Mythos highlights EU AI safety gaps, says researcher Caroli
Policy & Regulation
From Archives Deepseek's V4 Pro Update Responds to Flash, Holds Back Model Weights
From Archives Ling 3.0 Flash Cuts Hallucination Rate in Half for Open Models
From Archives Anthropic's USD 2 Trillion Valuation Fueled by USD 100 Billion Revenue Outlook
Business & Startups
From Archives Claude Cowork Brings AI Skills and Plugins Directly to Chrome Browser
LLMs & Generative AI
Common Questions Answered
What are the two releases DeepSeek launched and how do they differ?
DeepSeek released DeepSeek-V4-Pro, an updated flagship language model focused on agentic work with native support for OpenAI's Responses API and Codex integration, and DeepSeek Harness v0.1, an open-source agent harness released under the MIT license. While V4-Pro is the language model itself available through web interface, mobile app and API, Harness is the infrastructure layer that determines how models use tools, manipulate files, and execute long-running agent workflows.
How does DeepSeek Harness compete with Claude Code?
DeepSeek Harness moves competition beyond model intelligence and token prices into the agent tooling layer where Claude Code has become increasingly important. By releasing Harness as open-source under MIT license with code on GitHub, DeepSeek is now competing in the environment that determines how models manipulate files, maintain sessions, and execute complex agent workflows—territory previously dominated by Anthropic's Claude Code.
What is DeepSeek's strategic shift by releasing both V4-Pro and Harness together?
DeepSeek is copying Anthropic's playbook by controlling both the model and the environment it runs in, rather than just selling model weights and letting the ecosystem build tooling. This combined release allows DeepSeek to publish benchmarks that reflect both the model and scaffolding together, giving them more control over how their technology is perceived and used compared to competitors.
What platforms and integrations does DeepSeek-V4-Pro support?
DeepSeek-V4-Pro is live across the web interface, mobile app, and API with native support for OpenAI's Responses API and Codex integration. This multi-platform availability makes the model accessible to developers across different environments and workflows.
Further Reading
- DeepSeek Harness v0.1 enters developer preview with open-source code - CryptoBriefing
- DeepSeek unveils open-source harness, raises V4-Pro API prices - CryptoBriefing
- DeepSeek publicises efforts to challenge Anthropic's Claude Code - The Straits Times
- DeepSeek Is Building Its Own Claude Code. Beijing Wants the Whole Stack - Decrypt
- DeepSeek forms Harness team to rival Anthropic's Claude Code - Pandaily
