unfortunately Astra may dominate overall when it comes to cost/task instead of cost/token

https://x.com/cheatyyyy/status/2095597213896610184

The cost difference is honestly the most interesting part here.

If Gemini can get comparable results while costing 5–7x less, that changes the economics of running these models at scale. The fact that it’s thinking longer to get there is probably a feature, not a bug.

Google releases Gemini 3.8 Flash, its third Flash model in six weeks

i_am_a_seeker

The cost difference is honestly the most interesting part here.

If Gemini can get comparable results while costing 5–7x less, that changes the economics of running these models at scale. The fact that it’s thinking longer to get there is probably a feature, not a bug.