Related Story

    DeepSeek-R1's 671 Billion Parameters Model Outpaces OpenAI's o1 at 3% to 5% of Cost

    DeepSeek-R1's 671 Billion Parameters Model Outpaces OpenAI's o1 at 3% to 5% of Cost

    12 postsAI ModelingAI

    DeepSeek-R1, an open-source AI model developed by DeepSeek-AI, has emerged as a significant competitor to OpenAI's proprietary model, o1. With 671 billion parameters, DeepSeek-R1 was trained in around two months at a cost of $5.58 million, challenging the dominance of proprietary models by offering comparable performance at a fraction of the cost, reportedly at 3% to 5% of the expense of OpenAI's o1. The model's development focuses on enhancing reasoning capabilities through a unique multi-stage training process that includes reinforcement learning (RL) and a departure from traditional supervised fine-tuning (SFT). DeepSeek-R1 has demonstrated strong performance in various benchmarks, including mathematics, coding, and reasoning tasks, achieving scores such as 97.3% on the MATH-500 benchmark and an Elo rating of 2029 on Codeforces. It has been downloaded 109,000 times on HuggingFace, and its accompanying search feature is considered superior to competitors like OpenAI and Perplexity. In contrast, OpenAI's o1 models are known for their comprehensive safety measures and general capabilities across multiple applications, including creative writing and conversational AI. The competition between DeepSeek-R1 and OpenAI's o1 highlights the ongoing debate between open-source and proprietary AI development, with DeepSeek-R1's cost efficiency and accessibility challenging the high-cost paradigm of proprietary models.

    Proposed Markets