Text models
Estimate the cost of a Claude Sonnet 5.5 request
Measure one representative task and use billing records to calibrate the estimate.
Start from a real prompt
Measure input and output for a representative request. Count the full context sent in each turn of a conversation.
Compare reruns at the same quality
Test Sonnet and a lighter tier against the same output structure and accuracy targets. Check final billing records to correct the estimated output lengths.