Open-weight AI models spread fast; closed models still lead by ~4 months
The combination of a persistent capability gap and explosive distribution creates a policy problem: governments can control chip exports and data center locations but cannot remove model weights already in wide circulation. Friedberg argued that compute and intelligence infrastructure have become a form of national power, with cybersecurity named as a key arena where AI accelerates both offensive and defensive capabilities.
The full picture
Epoch AI found that since January 2026, the best open-weight models have trailed frontier closed models by an average of four months, measured as 8 points on its Epoch Capabilities Index, a gap it describes as comparable to the difference between GPT-5 and GPT-5.5. Against that backdrop, Moonshot AI published the full weights of its Kimi K3 model to Hugging Face on July 27, describing it as a 2.8-trillion-parameter model in a repository totaling about 1.56 terabytes, released under a custom license. David Friedberg argued that open-weight model downloads surged from 339 million in March 2025 to over 2 billion by March 2026, and that weights already distributed across thousands of computers are beyond the reach of government regulation, with Alibaba's Qwen alone approaching 1 billion downloads.
Sources
Related
Want this in your inbox?
I send one email each morning with the stories that moved. If you would rather just read here, that works too.
Subscribe free