On Wednesday, a Chinese artificial intelligence (AI) company introduced a new model that challenges OpenAI’s o1 AI in the realm of advanced reasoning. Named DeepSeek-R1-Lite-Preview, this large language model (LLM) boasts superior performance in several benchmark tests when compared to the o1 model. The model is available for free testing online, although users can only access its advanced reasoning feature a limited number of times. Moreover, the AI provides a transparent thought process, enabling users to understand the rationale behind its outputs.
DeepSeek-R1 AI Model Unveiled
Advanced reasoning represents a novel capability in LLMs, allowing them to engage in multi-step decision-making processes. This feature enables AI models to tackle complex inquiries that require a deeper contextual understanding and expert-level knowledge of specific subjects. Additionally, these models possess the ability to self-fact-check, which reduces the chances of generating inaccurate information.
Despite its advantages, advanced reasoning has not been widely adopted among foundational models. While some mixture-of-agent (MoE) models exhibit this capability, they consist of multiple smaller models working in tandem. In contrast, OpenAI’s o1 series has become well-known for its advanced reasoning features.
On Wednesday, DeepSeek announced the launch of the DeepSeek-R1-Lite-Preview model on X (formerly Twitter). The company asserts that its model surpasses the performance of the o1-preview on the AIME and MATH benchmarks, both of which assess an LLM’s reasoning and mathematical skills.
Testers from Gadgets 360 accessed the chatbot and observed that the AI reveals its complete thought process following each inquiry. This transparency allows users to trace the logical connections behind the model’s responses and identify any potential flaws. During their evaluation, testers noted the AI’s capability to tackle intricate questions.
The model’s quick response times also contributed to a seamless conversational experience. Currently, users can submit up to 50 messages to explore the “Deep Think” mode, which highlights the model’s reasoning process. Presently, it stands out as the only free AI model equipped with advanced reasoning features. Those interested can access the AI chatbot online here.
Furthermore, DeepSeek has announced plans to open-source the full version of the DeepSeek-R1 AI model in the near future, marking a potential milestone in the evolution of LLMs.
