Unverified50% confidenceFactExact time
每轮对话若Claude Code比Codex多输出30% Token，经过20轮对话累计差距可达11倍以上
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026
Sources
Codex vs Claude Code费用对比：10倍差价的真实原因拆解
bilibili赋范AI说6/9/2026
Related Claims
Unverified使用国产模型解决一个编程问题可能需要来回十来次对话，而使用Claude Code或Codex通常两三轮就能实现目标72% similarUnverifiedClaude Code 在多轮对话中会将历史消息累积传入模型，长时间工作后单次请求的输入 token 可能累积到数万甚至数十万70% similarUnverifiedClaude Code在SWE-bench上的多轮编码会话中，前缀缓存命中率平均约98%69% similarUnverifiedClaude Code consumes approximately 4x more tokens than Codex for equivalent coding tasks.68% similarUnverified通义千问Coder模型在Claude Code中两次对话就耗尽了37元人民币余额68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/43631API
curl https://kongchang.com/api/v1/knowledge/claims/43631MCP
get_claim(id=43631)