DeepSeek Launches Open-Source R1 Model to Rival OpenAI-o1
DeepSeek releases DeepSeek-R1, a highly capable open-source reasoning model that matches OpenAI-o1 performance. The model and its API are available now under a permissive MIT license.
DeepSeek officially releases DeepSeek-R1, a powerful new artificial intelligence model that achieves performance on par with OpenAI-o1. The fully open-source model and its accompanying technical report are now available to the public. Users can experience the model's advanced reasoning capabilities immediately by visiting the official DeepSeek chat interface.
The company makes a significant move for the AI industry by releasing the code and model weights under the permissive MIT License. This allows developers and businesses to freely distill and commercialize the technology. Additionally, DeepSeek provides six smaller distilled models based on DeepSeek-R1, with the 32B and 70B versions matching the performance of OpenAI-o1-mini.
DeepSeek-R1 relies on large-scale reinforcement learning during post-training to achieve a massive performance boost while using minimal labeled data. It excels particularly in math, code, and complex reasoning tasks. Developers can access the model through the API by setting the model parameter to "deepseek-reasoner," with highly competitive pricing that includes steep discounts for cache hits.