Chinese AI startup DeepSeek has made a significant breakthrough with its new open-source AI model, DeepSeek-R1, which has been shown to outperform OpenAI's o1 in several math, coding, and reasoning benchmarks.¹ This achievement is notable not only because of DeepSeek-R1's impressive performance but also due to its significantly lower cost.
DeepSeek-R1 was trained using a combination of reinforcement learning and supervised fine-tuning, which enabled it to achieve state-of-the-art performance at a fraction of the cost of similar models.² In fact, DeepSeek-R1's training cost was around $5.5 million, compared to OpenAI's estimated $100 million spent on training GPT-4.³
The implications of DeepSeek-R1's success are significant, as it demonstrates that high-performance AI models can be developed at a lower cost and with greater accessibility. This could potentially democratize access to AI technology and enable more researchers and developers to contribute to the field.
