DeepSeek releases R1, a reasoning AI model that matches top competitors' capabilities while using significantly fewer computing resources.

What happened: DeepSeek introduced R1, an open-weights reasoning model (an AI that shows its step-by-step thinking before answering). The company claims R1 achieves performance comparable to OpenAI's o1 while using 27% of FLOPs compared with DeepSeek-V3.2 and requiring 83.9 GiB of memory.

Why it matters: The release demonstrates that high-performance reasoning AI can be built far more efficiently than previously shown, challenging assumptions about the computing scale needed to compete with leading closed systems. This may shift how companies think about the computational investment required for advanced AI capabilities.

What to watch: R1 is available as an open-weights model, meaning researchers and developers can inspect and modify it. The company also offers a closed-source version accessible through its API, giving different audiences options for how they use the technology.