DeepSeek releases R1, an open-weights reasoning model that ranks #2 globally and uses 27% of the compute of comparable competitors.

What happened: DeepSeek has released R1, an open-weights reasoning model (an AI that shows its working steps before answering). The model ranks as #2 among open-weights reasoning models globally. It achieves this performance while using 27% of FLOPs (computational operations) compared with DeepSeek-V3.2.

Why it matters: Open-weights models let researchers and companies inspect and modify the underlying AI system, rather than rely on a closed commercial product. A top-tier reasoning model available in this transparent form shifts power toward developers outside the largest AI labs. The dramatic efficiency gain suggests reasoning performance is not locked behind massive compute budgets.

What to watch: R1 is available now as an open-source release. A hosted commercial version called DeepSeek Cloud is coming soon.