AI

This is where we track what frontier systems can actually do, not what a launch post says they will. Model releases, agent architectures, training and inference economics, open-weight releases, and deployments that survived contact with production.

We skip benchmark victory laps with no methodology and demos that never ship. When a claim comes with a number, we carry the number.