Look At These Ridiculous Numbers

Two months of Claude Code usage and what it would’ve cost me at API rates

I ran out of Claude Code credits and fell down a rabbit hole. Here’s the damage.

What I Actually Used (8 weeks)

This is raw token consumption across models. Note: 92% of those tokens are “cache reads” which I’ll explain in a sec.

Model	Input	Output	Cache Read	Cache Write	Total
Opus 4.5	950,860	2,588,991	1,358,438,079	122,308,882	1,484,286,812
Sonnet 4.5	201,228	4,151,480	684,762,875	50,804,886	739,920,469
Opus 4.1	708	67,469	6,475,584	464,816	7,008,577
Haiku 4.5	96	139	266,580	67,004	333,819
TOTAL	1,152,892	6,808,079	2,049,943,118	173,645,588	2,231,549,677

Yes, 2.2 billion tokens. But here’s the thing: every message you send in Claude Code includes the ENTIRE conversation history. So the context gets re-sent constantly. Prompt caching means if it’s already been sent, you pay 10% of normal price. That’s what “cache read” means. It’s how this doesn’t bankrupt everyone.

What That Would’ve Cost at API Rates

Using Anthropic’s published pricing (January 2025):

Model	Input	Output	Cache Read	Cache Write	Total
Opus 4.5	$14.26	$194.17	$2,037.66	$2,293.29	$4,539.39
Sonnet 4.5	$0.60	$62.27	$205.43	$190.52	$458.82
Opus 4.1	$0.01	$5.06	$9.71	$8.72	$23.50
Haiku 4.5	$0.00	$0.00	$0.02	$0.07	$0.09
TOTAL	$14.87	$261.51	$2,252.82	$2,492.59	$5,021.80

I paid $200 (two months of the $100 Max plan). API equivalent: $5,021.80. That’s a 25x value multiplier.

If You Maxed Out the Subscription

This is the wild part. Anthropic publishes weekly usage limits. If you used every single hour they allow:

Plan	Monthly Cost	Maxed API Equivalent	Value Multiplier
Max 5x	$100	~$20,750	~208x
Max 20x	$200	~$33,500	~168x

I’m apparently using about 12% of what the plan allows. Someone grinding harder than me is getting 200x value.

Anthropic is either losing money on power users, betting inference costs drop fast, or both. Either way: good deal.