This post is also available in:
עברית (Hebrew)
DeepSeek, a rising name in AI, has unveiled its new open-source large language model, the DeepSeek-R1, claiming it rivals top models like OpenAI’s o1-1217 in multiple benchmarks. With 671 billion parameters and a unique approach to activation—only 37 billion are engaged at any given time—DeepSeek-R1 promises exceptional performance while maintaining computational efficiency. The model is already receiving praise for its advanced capabilities in tasks like mathematical reasoning, coding, and general language understanding.
DeepSeek-R1 reportedly outperforms its competitors in several crucial areas. According to internal tests, it scored 79.8% on AIME 2024, slightly above OpenAI’s o1-1217. On MATH-500, it achieved a 97.3% score, also slightly better than OpenAI’s model, while outperforming others. The model also excels on coding platforms like Codeforces, putting it in the top 4% of global participants. Its open-source nature allows developers and companies to download and implement the model, a significant advantage for those looking to avoid reliance on proprietary AI services.
In terms of pricing, DeepSeek-R1 is very competitive, offering an API priced far cheaper than OpenAI’s offerings. Furthermore, smaller models are also available for those with less demanding needs.
However, while the model’s technical performance has garnered attention, its censorship features have raised eyebrows. According to a report by Cybernews, DeepSeek-R1 refuses to address sensitive topics like the Tiananmen Square protests, Taiwan, or Chinese cyber threat actors, often redirecting such queries with vague disclaimers. The model adheres strictly to a narrative aligned with the Chinese government’s stance. This censorship is a key point of contention, as it limits the model’s objectivity and undermines its global appeal.
Despite these concerns, DeepSeek-R1’s technical specifications and open-source availability position it as a formidable player in the AI landscape, potentially disrupting the dominance of US tech giants. However, its censorship policies may hinder its adoption in markets where freedom of information is a priority, and maybe it’s better this way.
