DeepSeek Launches 'Harness' Team for AI Agents, Unveils V4 Pro to Rival Claude Code
SingTao · 2 SOURCESabout 2 hours ago2 MIN
Summary
Chinese AI startup DeepSeek has established a dedicated team called "Harness" on WeChat to develop AI agents capable of handling programming and complex workflows, directly challenging Anthropic's Claude Code. The company also released the official version of its flagship V4 Pro model, which supports a 1 million token context window and demonstrates performance comparable to Fable 5 in multiple benchmarks. The WeChat account is owned by a Beijing-based entity controlled by DeepSeek and has received verification from Tencent (stock code: 700) .
Key Points
- DeepSeek created the "DeepSeek Harness Team" public account on WeChat to recruit talent for developing AI agent products that can handle programming and complex workflows
- The V4 Pro official version supports 1 million token context and up to 384,000 token output, with pricing at 0.025 RMB per million input tokens (cache hit), 3 RMB (cache miss), and 6 RMB per million output tokens
- The "Harness" technology adds tool calling, task planning, memory management, and result verification to large language models, enabling multi-step execution beyond single-question responses
- Evaluation results show V4 Pro's performance approaches Anthropic's Fable 5 in multiple tests, significantly improving over earlier preview versions
- DeepSeek's recruitment materials indicate the team will focus on transforming their models into frontier AI agent products for software development and professional work
Why It Matters
DeepSeek's expansion into AI agents marks a significant escalation in the global competition among AI developers, with the company positioning itself to challenge Anthropic's dominance in AI-powered coding tools . The V4 Pro's competitive pricing combined with Fable 5-level performance could make advanced AI agent capabilities more accessible to developers worldwide, potentially disrupting the current market landscape .
DeepSeek's expansion into AI agents marks a significant escalation in the global competition among AI developers, with the company positioning itself to challenge Anthropic's dominance in AI-powered coding tools . The V4 Pro's competitive pricing combined with Fable 5-level performance could make advanced AI agent capabilities more accessible to developers worldwide, potentially disrupting the current market landscape .