DeepSeek launches AI model DeepSeek-R1 to compete with OpenAI's o1

Chinese AI company DeepSeek has launched DeepSeek-R1, a new reasoning model that reportedly matches OpenAI's latest model, o1. This model fact-checks itself, aiming to reduce common AI errors, but can take over 10 seconds to process tasks. The current version, DeepSeek-R1-Lite-Preview, performs well on AIME and MATH benchmarks. However, it still has limitations, such as difficulties with logic problems and vulnerabilities that allow users to bypass restrictions. DeepSeek-R1 avoids politically sensitive topics, likely due to Chinese government regulations. The company plans to release the model as open-source and will provide an API in the future.