It's possible that prices will go up (although the cost of pure inference tends to go down; the question is more how to amortize the cost of training new models).

But this is absurd:

> the cheapest usable tier of Claude Code is $100/mo

If you pay by the token instead of with a subscription, and don't send the entirety of your code base with each request, costs are ridiculously low. Like, $50 will last a minimum of 3 months of heavy use on openrouter.

It's also far from certain everyone needs the latest version of the best "frontier model"; it very much depends on what you do.

$50 will last you a good long time if you don't use many tokens, and you are judicious about which model you use, and don't need web search much.

However, on a fixed price plan your behavior changes. It's a qualitative change in how you work, rather than quantitative. Ideation and product design and specification start becoming bottlenecks.

I started out the API route. I started spending $100 a month once I was spending upawards of $10 in tokens a session.

I am imagining working for a company in the future where prompt reviews are required because the company is cheap.