2025 was a turning point for open-source AI, and Nathan Lambert just published the receipts.

His year-end review tracks 600+ model releases and lays out who's actually leading: DeepSeek and Qwen at frontier tier, Meta demoted to honorable mentions, and Chinese labs dominating the landscape.

The takeaway? Open models aren't the "privacy compromise" option anymore. They're genuinely competitive.

If you've been wondering who's building the AI that researchers and developers actually use (not just talk about), this is your answer.

https://www.interconnects.ai/p/2025-open-models-year-in-review