On August 19, 2025, DeepSeek, a leading Chinese AI company, announced an enhanced version of its V3 artificial intelligence model, named V3-0324, now available for testing. The update, shared via Hugging Face, promises significant improvements in reasoning, coding, and performance, positioning it as a strong competitor in the global AI landscape.
The V3-0324 model retains the Mixture-of-Experts (MoE) architecture with 685 billion parameters, activating 37 billion per token for efficiency. Notable enhancements include:
- Improved Reasoning: A 20-point boost on the AIME math benchmark, showcasing advanced problem-solving capabilities.
- Enhanced Coding: Stronger front-end development skills, ideal for web development tasks.
- Multi-Token Prediction (MTP): This feature predicts multiple tokens simultaneously, improving accuracy and inference speed.
DeepSeek recommends using V3-0324 for non-complex tasks with the “DeepThink” mode disabled, while its R1 model remains superior for advanced reasoning.
DeepSeek announced what appeared to be an update to its older V3 artificial intelligence model on Tuesday, declaring an enhanced version ready for testing https://t.co/O9JXVomXhQ
— Bloomberg (@business) August 19, 2025The updated model is accessible via Hugging Face and DeepSeek’s website, with an OpenAI-compatible API for developers. It supports local deployment on NVIDIA and AMD GPUs, as well as Huawei Ascend NPUs, with a 128K context window for handling complex tasks. DeepSeek encourages community feedback to refine the model further.
The V3-0324 release follows DeepSeek’s pattern of impactful, open-source AI advancements, challenging industry leaders like OpenAI. Despite concerns about security vulnerabilities in earlier models, the update is licensed under the MIT License for commercial use, potentially reshaping AI development in China and beyond.
Speculation on platforms like Reddit suggests V3-0324 may pave the way for DeepSeek’s anticipated R2 model, expected by late 2025. As the company continues to innovate, this update reinforces its commitment to open-source AI and competitive performance.
