DeepSeek, a Chinese AI lab, has introduced DeepSeek-R1, a reasoning-based AI model that rivals OpenAI’s o1 on benchmarks like AIME, MATH-500, and SWE-bench Verified. Available on Hugging Face under an MIT license, R1 is scalable, with versions ranging from 1.5 billion to 671 billion parameters—the latter being among the largest models to date.

The model’s focus on reasoning enables self-fact-checking, making it more reliable in domains like math, science, and programming. While slower than non-reasoning models, R1’s accuracy and affordability—offered at 90%-95% lower API costs than OpenAI’s o1—are major draws. Despite its advancements, R1 faces limitations due to Chinese regulatory filters on politically sensitive topics.

The launch of R1 amidst rising U.S.-China tensions in AI, with stricter export rules potentially limiting China’s access to advanced semiconductor technology. As Chinese labs like DeepSeek, Alibaba, and Kimi push forward, industry experts predict further competition in the global AI landscape.

Source: TechCrunch

For more details, read the full article here.