Open-weight language models have improved fast, and for a growing set of tasks the gap with the best closed models is narrowing. For developers and privacy-conscious teams, that means viable options you can run on your own hardware without sending data anywhere.
The trade-offs still matter: closed frontier models lead on the hardest reasoning, while open models win on cost, control and customization. We track the releases worth your attention and what they're actually good for.