GB
Gaurav Bhargava
AI Cost Intelligence
Your AI Bill Is High.
Here Is Exactly Why.
Tokens are the unit. Output costs 5× more than input. Context re-sends on every call. Here's what to do about it.
1 token ≈
4 chars
or ¾ of a word
1,000 words ≈
1,300
tokens sent to model
You are charged
Twice
input tokens + output tokens
Current pricing — per million tokens
Haiku 4.5
Input
$1
Output
$5
Fast, high-volume tasks
Sonnet 4.6
Input
$3
Output
$15
Best price-to-quality ratio
Opus 4.7
Input
$5
Output
$25
Complex reasoning tasks
5×
Output tokens cost 5× more than input — across every model.
The model thinks by generating text. Reasoning costs more.
Three levers that actually reduce your bill
90%
Prompt Caching
Cache stable system prompts. Subsequent calls read from cache at 10% of the input price.
Instant saving
50%
Batch Processing
Async workloads don't need real-time responses. Batch API halves every token — no quality difference.
Async workloads
25×
Model Routing
Route classification and extraction to Haiku. Reserve Sonnet or Opus for reasoning that needs it.
Biggest variable
Follow for more
Gaurav Bhargava
@YourGauravB
gauravbhargava.ai
Enterprise AI · Cost Intelligence · FinOps