OpenAI’s O1 Performance Surpassed by DeepSeek’s R1-Lite-Preview AI Model
DeepSeek is an AI subsidiary primarily focused on high-performance open-source technology, emerging from High-Flyer Capital Management in China. Currently, DeepSeek is testing the R1-Lite-Preview, a reasoning-focused large language model (LLM) for potential use.
In some test scenarios, the R1-Lite-Preview has demonstrated performance comparable to OpenAI’s O1-preview model, and in some cases, even superior capabilities. A notable feature is that the R1-Lite-Preview model, tested in September 2024, showcased “chain-of-thought” reasoning abilities, capable of providing detailed, step-by-step explanations for user-submitted data and questions. It can explain not just what happened, but why something occurred.
While some explanations might seem meaningless to human users, the DeepSeek-R1-Lite-Preview model has been trained to answer even indirect questions precisely. These seemingly irrelevant responses are potential results of its comprehensive training. As examples, it demonstrated precise answers to questions like “how many letter Rs are in the word Strawberry?” and “which is larger 9.11 or 9.9?”
Users can access R1-Lite-Preview through DeepSeek Chat at chat.deepseek.com. While the service is free, the “Deep Think” advanced model is limited to 50 messages per day.
