Key facts
- DeepSeek launched the beta versions of its V4 AI models, including V4-Pro and V4-Flash.
- The new models boast enhanced knowledge, reasoning, and agentic capabilities.
- DeepSeek's V4 models utilize Huawei chips, decreasing dependence on U.S. suppliers.
- The V4-Pro model is priced substantially below Western AI offerings.
- DeepSeek claims its V4-Pro Max outperforms OpenAI's GPT-5.2 and Google's Gemini 3.0-Pro on reasoning benchmarks.
DeepSeek, a Chinese artificial intelligence company, has released the beta versions of its latest V4 models, intensifying the global price war in advanced AI. The V4-Pro and V4-Flash models, launched on Friday, offer significant improvements in knowledge, reasoning, and agentic capabilities. These new models are partly supported by computer chips from Chinese tech giant Huawei, reducing DeepSeek's reliance on U.S. chipmakers like Nvidia.
The V4 models come with one million token context windows and are priced considerably lower than their Western counterparts. DeepSeek claims its V4-Pro Max version demonstrates superior performance on standard reasoning benchmarks compared to OpenAI's GPT-5.2 and Google's Gemini 3.0-Pro, though it falls slightly short of GPT-5.4 and Gemini 3.1-Pro. This release follows OpenAI's launch of its GPT-5.5 model just hours prior.
The V4-Pro model features 1.6 trillion total parameters, making it the largest open-source model currently available. It utilizes a Mixture-of-Experts approach, activating only a relevant slice of its parameters per request to enhance efficiency. DeepSeek's previous specialized reasoning model, R1, released in January 2025, had previously surprised markets with its cost-effectiveness.