Pinned
The best model is a staffing decision, not a leaderboard result.
We ran 14 models on real work. Costs varied 50x. The benchmark winner missed the podium. No model won every job.
Choose the job. Set the bar. Run the lightest model that reliably clears it.




