November 2023: DeepSeek-Coder Release
DeepSeek introduced DeepSeek-Coder, an open-source model ailable under the MIT license. This model was designed to facilitate code generation and understanding, catering to both researchers and commercial users.
May 2024: DeepSeek-V2 Series Launch
The company unveiled the DeepSeek-V2 series, comprising base models and chatbot variants. Notably, DeepSeek-V2 achieved a cost of 2 RMB per million tokens, offering a more affordable alternative to existing models. It secured the seventh position in the Tiger Lab’s LLM ranking.
December 2024: Introduction of DeepSeek-V3
DeepSeek released DeepSeek-V3, a model with 671 billion parameters, trained in approximately 55 days at a cost of $5.58 million. This model outperformed competitors like Llama 3.1 and Qwen 2.5, matching the performance of models such as GPT-4o and Claude 3.5 Sonnet.
January 2025: Launch of DeepSeek-R1 and Chatbot Application
DeepSeek introduced DeepSeek-R1, a model optimized for logical inference and real-time problem-solving. Alongside this, the company released a chatbot application based on DeepSeek-R1, which gained significant traction, surpassing ChatGPT in free app rankings on the iOS App Store in the U.S.