On IBM’s Q3 earnings call, CEO Arvind Krishna made the case that open-weight models now make total sense for running AI on your own hardware.
What he said
- In 2024 a frontier model might score 70 on a broad benchmark while open models scored 20 or 30. By late 2025 it was more like 70 to 50.
- Today open models run at “anywhere from 25 percent to 10 percent of the token cost” of frontier models while being “90 percent as effective.”
- “Would you like something that’s 10 percent of the cost if it’s 90 percent as good as the best?” He thinks most businesses will pick the cheaper option.
- The pitch: that kind of capacity fits in two or three racks.
My take
This is exactly why I keep playing with local AI. When a big enterprise vendor says the gap has closed enough, the self-hosting crowd isn’t the fringe anymore.
Have a blessed day! And Happy Hacking!
