Opus 5.5 pricing cuts real Claude Code session costs by 31%
The sticker-price drop understates the savings — billed cost is driven by turn count and cache reuse, so /clear, /compact timing and MCP toggling matter more than the per-token rate. Developers running Claude Code should audit their reasoning effort settings too: Opus 5.5 defaults to medium and reasons more per level, so prompts tuned for Opus 5 need re-tuning. Route subagents for search and log reading to Sonnet or Haiku to keep Opus spend on code writing.
Opus 5.5 is 20% cheaper per input and output token than Opus 5, and 60% cheaper on cache reads.
Because each Claude Code turn resends the full conversation, total session cost depends on how many rounds a task runs, and the new pricing works out to roughly 31% savings per session.
Anthropic's internal test moving 44 customer support tickets from Opus 4.8 to Opus 5.
5 cut costs about 18%, and about 25% after a prompt review.Subscription users on Pro, Max and Team plans get around 25% more usage at the new rate.
Cache expires after 5 minutes on the API but 1 hour for subscribers, and a 120,000-token context round can jump from US$0.
02 to US$0.60 once the cache lapses.
这篇文章解释清楚了为什么现在 Opus 5.5 要比之前 Opus 5 耐用多了，看起来 API 价格降低了 20%（缓存读取降了 60%），但是账单高低其实看任务跑了多少轮，因为每跑一轮，都要把之前的整段对话重新发给模型一遍。 这样算下来，平均一个会话能便宜约 31%。 Anthropic 内部用 44 张客服工单测试，从 Opus 4.8 换到 Opus 5.5 成本降了约 18%，再做一次提示词审查，总共比原来低了约 25%。 订阅用户（Pro、Max、Team）的额度按新价格折算，能比用 Opus 5 时多撑 25% 左右。 一些技巧： 1. 换到不相关的任务，用 /clear 清空对话。 2. 想保留前情就用 /compact 压缩，但要在缓存失效之前做，否则要贵一些。 3. 订阅用户的缓存保留 1 小时，用 API 的话，缓存只保留 5 分钟，离开 6 分钟，12 万 Token 上下文的下一轮就从 0.02 美元涨到 0.60 美元。 4. 缓存只复用和上一轮开头完全一致的部分，中途切换模型、接入或断开 MCP 服务器、第一次打开快速模式，下一轮都要按更贵的写入价重新付费。 5. Opus 5.5 默认推理强度是 medium，比 Opus 5 的默认 high 低一档。但同一档位下 Opus 5.5 想得更多，所以给 Opus 5 调好的档位别照搬。日常活用 medium，卡住了调 high，批量重命名这类机械活用 low。 6. 模型分工上，日常用 Opus 5.5。开到 xhigh 还在同一个问题上连卡两次，再换更贵的 Fable 5.1，解决完切回来。搜代码、读日志的 subagent 交给 Sonnet 或 Haiku，写代码留在 Opus 5.5。 ClaudeDevs @ClaudeDevs Opus 5.5 is 20% cheaper per input and output token than Opus 5, and 60% cheaper on cache reads. So what does that actually do to the cost of a task in Claude Code? We ran the numbers, and built a calculator so you can run yours from /usage: claude.dev/blog/what-a-ta… 🔗 View Quoted Tweet
This report is categorized under AITOP's "How-to" desk, focusing on US frontier AI labs and Silicon Valley technology leaders. Within our cross-regional analytical framework, it directly maps to AI developer toolchains, coding agents, and engineering workflows and US-China foundation model benchmarks and commercialization comparison. Key entities and technical topics under continuous tracking include Claude, Opus 5.5, Anthropic, prompt engineering.
Source attribution: curated from 宝玉 (@dotey). Cross-source corroboration: 11 independent sources have covered this development (including claudedevs, IT之家, Riley Goodside), confirming industry-wide signal strength.
