The Chinese lab DeepSeek has unveiled a new open version of its DeepSeek-R1 model, which, according to them, can compete with OpenAI o1 on certain AI benchmarks. The R1 model is available on the Hugging Face platform under the MIT license, allowing unrestricted commercial use. DeepSeek claims that R1 outperforms o1 on the AIME, MATH-500, and SWE-bench Verified tests. The R1 model effectively verifies its own results, helping to avoid errors that often occur in other models.
🚀 DeepSeek-R1 is here!
— DeepSeek (@deepseek_ai) January 20, 2025
⚡ Performance on par with OpenAI-o1
📖 Fully open-source model & technical report
🏆 MIT licensed: Distill & commercialize freely!
🌐 Website & API are live now! Try DeepThink at https://t.co/v1TFy7LHNy today!
🐋 1/n pic.twitter.com/7BlpWAPu6y
DeepSeek-R1 contains 671 billion parameters, making it one of the largest models in the world. At the same time, DeepSeek has released “distilled” versions of R1, ranging from 1.5 billion to 70 billion parameters, allowing the smallest ones to run on regular laptops. The full version of R1 requires more powerful hardware, but is available via the DeepSeek API at prices 90–95% lower than those of OpenAI o1.
The R1 model is subject to regulation by Chinese internet authorities, meaning it will not respond to queries that may displease regulators, such as questions about the Tiananmen Square events or Taiwan’s autonomy. This restriction is common among many Chinese AI models, which avoid answering sensitive topics.
