Unverified50% confidenceFactExact time
Cost per task = input tokens × input unit price + output tokens (including thinking tokens) × output unit price.
1
Sources
50%
Confidence
Long-term
Relevance
10/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedClaude Code's API billing typically prices input tokens and output tokens separately; the linear accumulation of input tokens across conversation turns is the primary driver of cost inflation in long tasks.79% similarVerifiedEvaluating the actual cost of an LLM should use 'cost per task' rather than 'cost per token'—that is, the total token consumption required to complete a unit task multiplied by the unit price.77% similarVerifiedThe output token unit price of large models is generally higher than that of input tokens.76% similarUnverifiedPer-request billing is flat and predictable; per-token billing is a function of input, output, and the number of loop retries, so the cost shape is different.75% similarUnverifiedCost accounting should separately track the volume and cost of input tokens and output tokens, rather than looking at total token count in aggregate.74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1059543API
curl https://kongchang.com/api/v1/knowledge/claims/1059543MCP
get_claim(id=1059543)