DeepSeek V4 AI models set new benchmark in open-source

DeepSeek launched V4-Pro and V4-Flash AI models with up to 1.6T parameters and a 1M token context window, offering high efficiency and performance, challenging industry leaders.

DeepSeek has unveiled its V4 series of open-source AI models, featuring two variants: V4-Pro, with approximately 1.6 trillion parameters, and V4-Flash, with 284 billion parameters. Both models support a 1 million token context window, matching leading solutions like Google’s Gemini. V4-Pro is reported to deliver performance on par with top closed-source systems, excelling in reasoning, coding, and STEM tasks. V4-Flash offers a more efficient and cost-effective alternative, making advanced AI accessible to a broader audience. DeepSeek highlights significant reductions in computational and memory requirements compared to competitors. The models are released under the MIT license, with weights available on Hugging Face and ModelScope. This launch coincides with reports of major Chinese tech firms considering investments in DeepSeek, potentially valuing the company above $20 billion. While the V4 models are currently text-only, future developments and deployment on Huawei chips could further strengthen China’s AI ecosystem. Real-world performance and scalability will be crucial for widespread adoption.