DeepSeek-R1DeepSeek-R1

In January 2025, the Chinese company DeepSeek released R1, an open-weight reasoning model that matched the performance of leading systems from far larger and better-funded American labs. It shipped freely, alongside a detailed technical report.

What made R1 remarkable was how it learned to reason. Rather than relying heavily on human-written examples, DeepSeek used large-scale reinforcement learning to let the model discover effective step-by-step problem solving on its own, rewarding it for reaching correct answers.

The model produced long internal chains of thought, working through math and coding problems much as a person might sketch on scratch paper before answering. It rivaled OpenAI's o1 on many reasoning benchmarks at a reported fraction of the training cost.

The release sent a shock through financial markets. Investors who had assumed that frontier AI required enormous, exclusive budgets suddenly reconsidered, and chipmaker Nvidia alone shed hundreds of billions of dollars in market value in a single day.

Beyond the market drama, R1 was a milestone for open, reproducible reasoning models, and a sign that the frontier of AI was more contested, and more global, than many had assumed.