China-based AI startup DeepSeek has introduced an open-source version of its reasoning model, DeepSeek-R1, claiming that it performs comparably to OpenAI’s o1 on various AI benchmarks.
DeepSeek-R1 tackles problem-solving by emulating human reasoning. Developed by DeepSeek, which recently released the open-source DeepSeek-V3, this advanced model enhances the lab’s reputation for delivering high-performance AI systems at a lower cost than competitors like Meta and OpenAI.
This new model is aimed at improving the problem-solving and analytical skills of AI systems, and features two main versions: DeepSeek-R1-Zero and DeepSeek-R1.
The DeepSeek-R1-Zero version is developed solely through reinforcement learning (RL) and does not involve any supervised fine-tuning. In contrast, the DeepSeek-R1 builds upon the groundwork established by R1-Zero. It includes a cold-start phase that utilizes carefully selected data and multi-stage RL, which significantly boosts its reasoning abilities and overall clarity.
R1 can be accessed on the AI development platform Hugging Face under an MIT license, allowing for unrestricted commercial use. DeepSeek claims that R1 outperforms o1 on benchmarks such as AIME, MATH-500, and SWE-bench Verified. AIME uses other models to assess performance, MATH-500 consists of various word problems, and SWE-bench Verified is centered on programming tasks.
It should also be mentioned that R1 has 671 billion parameters, as disclosed by DeepSeek in a technical report. These parameters are indicative of a model’s ability to solve problems, and typically, models with a higher number of parameters tend to outperform those with fewer.
As a reasoning model, R1 is capable of self-fact-checking, which helps it sidestep common errors that often affect other models. While reasoning models typically take longer—ranging from seconds to minutes—to reach conclusions compared to standard models, they generally offer greater reliability in fields like physics, science, and mathematics.
With its ability to tackle intricate mathematical and reasoning challenges, DeepSeek-R1 holds promise for educational and tutoring applications. Its coding skills also make it a potential asset for software development tasks, including code generation and debugging. Additionally, its strength in long-context comprehension and question-answering makes it a valuable tool for research purposes.
- Meta’s AI Video Editor Reimagines Short Videos Instantly - June 13, 2025
- Google Search Gets Smarter with Real-Time Data Visualisation Tools - June 9, 2025
- Bing’s New AI Video Creator Uses OpenAI’s Sora – And It’s Free - June 3, 2025
