DeepSeek, a Chinese AI startup, launched its LLM R1 on January 10, 2025, claiming it rivals OpenAI’s ChatGPT for reasoning tasks while costing under $6 million to train. Founded in May 2023, it has gained significant attention after reaching top App Store charts and stirring stock market reactions. DeepSeek offers models based on Llama and Qwen with both general and reasoning-focused versions, while their architecture optimizes for limited hardware capabilities, such as the NVIDIA H800 GPUs. Initial usage shows promising multilingual performance but raises questions around security, privacy, and training data transparency. Reproduction of reported performance remains pending as researchers attempt to validate results.
https://www.thoughtworks.com/insights/blog/generative-ai/demystifying-deepseek