Chinese AI lab DeepSeek has released R2, an open-source reasoning model that matches Anthropic's Claude Opus 4.6 on several key benchmarks — a significant achievement for the open-source AI movement.
The model is available for free download under an Apache 2.0 licence, allowing anyone to use, modify, and deploy it without restrictions. This represents a major challenge to the business models of closed-source AI providers.
DeepSeek R2 scored 49.3% on Humanity's Last Exam, compared to Claude Opus 4.6's 50.1% — within the margin of error. On coding benchmarks, R2 actually outperformed Claude on several tests.
The release has reignited debate about the sustainability of the closed-source AI business model. If comparable performance can be achieved for free, how will companies like OpenAI and Anthropic justify their premium pricing?
The geopolitical implications are also significant. DeepSeek's success demonstrates that China can produce frontier AI models despite US export controls on advanced chips. The lab reportedly achieved these results using older Nvidia A100 chips.