Quick model indicator overhaul

Suggestion is 2 fold:
Give a better idea of actual usage
Give a super quick idea of intelligence

Goal of it is to hopefully have more people give cheap models like k2.5 a try or even use non-think models like 5.2

For accuracy sake an indicator based on cost to run would be more accurate for actual model usage compared to token cost; maybe use AA numbers for this if available

This would lead more people using non thinking models for example.

In the same line show a “smartness indicator” for the intelligence of a model based on an avr on the benchmarks