News Summary

DeepSeek, a Chinese AI firm, has unveiled its latest AI model, DeepSeek V3, which is considered one of the most potent ‘open’ AI models available. Released under a permissive license, DeepSeek V3 allows for modifications and commercial use, making it accessible for a broad range of applications including coding, translating, and writing tasks. The model has outperformed both openly available and closed AI models in various benchmarks, including coding competitions hosted on Codeforces, surpassing notable models like Meta’s Llama 3.1 and OpenAI’s GPT-4o. DeepSeek V3 boasts an impressive training dataset of 14.8 trillion tokens and contains 671 billion parameters, significantly larger than its competitors. Despite the model’s massive size and capabilities, it was developed on a relatively modest budget of $5.5 million using 2048 Nvidia H800 GPUs over two months. However, the model’s responses are influenced by Chinese regulations, avoiding politically sensitive topics. DeepSeek is backed by High-Flyer Capital Management, aiming to achieve superintelligent AI, and has made substantial investments in server clusters for AI development.