The Deepseek R1 AI model is said to deliver powerful performance for only three to five percent of the cost of OpenAI’s models.
Deepseek has made the AI community rethink their strategies and what it takes to achieve advanced AI performance. The open source model rivals the performance of OpenAI’s o1 for only three to five percent of the cost, VentureBeat knows. So are the investments of the big AI players necessary?
Far from market leader, but groundbreaking
The model launched Monday and immediately became the most trending model on HuggingFace with 131,000 downloads at the time of writing. Developers want to find out what the model means for their AI development. DeepSeek’s accompanying search function is said to work better than those of OpenAI and Perplexity, matched only by Google’s Gemini Deep Research.
What makes DeepSeek unique is its focus on reinforcement learning (RL) rather than the commonly used supervised fine-tuning (SFT). By bypassing this process, DeepSeek has created a model that reasons independently and prioritizes more complex problems. Deepseek reportedly achieved that performance with a budget of $5.58 million, which is vastly less than the millions invested by OpenAI and Meta.
Its low cost and open access make the R1 model a perfect alternative to more expensive competing models. The release of the advanced model could lead to democratization, effectively allowing smaller organizations to compete.