Why newer models not reducing the less token and more output?

This may sound not right or diabolical, but I think the whole point to release a newer advanced model is to take less tokens and give more better outcomes than the previous ones.

I was using SOL 5.6 as a plus member, switched to astra yesterday and started using it in my project. I believed that it would make my work even faster, as I watched the overhyped launch videos on YouTube.

And here I am, it just runs 3-5 minutes and makes some edits then just eats over 20% of weekly and 100% of the 5 hour limit.

That's not what I expected.