Unverified50% confidenceOpinionExact time
In a Hacker News discussion, an engineer pointed out that the cost of a Claude Code session is overwhelmingly dominated by the cost of cached input tokens, and what really drives up expenses is repeated context reading, not model choice itself.
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
10/17/2026
First Seen
Valid until: 10/31/2026
Sources
Related Entities
Related Claims
UnverifiedClaude Code's official cost documentation acknowledges that /compact reads the conversation to be summarized, so compacting a large context is itself a large request, but it does not give a specific cost figure.72% similarUnverifiedClaude Code作为命令行工具，每次交互需要把大量代码库上下文传入，导致token消耗更快68% similarVerifiedIn automated coding loop scenarios with Claude Code, a single task may consume hundreds of thousands or even millions of tokens.68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1051009API
curl https://kongchang.com/api/v1/knowledge/claims/1051009MCP
get_claim(id=1051009)