This is spot on. My co-workers are burning through token allotments mid-month. The economics to get to the final session output is what's paramount. It's refreshing to see something tailored to coding economically. But I suspect even with this, the drive will be to having a master agent control the loop where it assess for each task what is the cheapest model likely to do the task accurately
2026-08-15 20:22:02
1
Lorand Veres :
I haven't tried the 3.7 yet, but I used the flash and pro 3.5 and 3.6. For everyday work the flash outperforms the Pro. I used the pro for deep planing/research and blueprints. The pro for small tasks it failed to find little small bugs vs the flash what did well on that tasks. Of course I haven't used them for media generation, and can't tell their performance on that.
2026-08-15 14:27:37
1
Agentic Engineering :
👋👋👋👋
2026-08-14 14:30:26
0
To see more videos from user @agenticengineering, please go to the Tikwm
homepage.