Opus 5.5 list prices drop 20% on input and output and 60% on cache reads — Anthropic’s same-token session falls from $3.50 to $2.40

What changed

ClaudeDevs posted that Opus 5.5 is 20% cheaper than Opus 5 on input and output tokens and 60% cheaper on cache reads, and linked a calculator on Anthropic’s claude.dev blog. The blog states every Opus 5.5 price line is lower, with cache reads moving from one-tenth of the input price to one-twentieth. Examples use API list prices of $4 per million input tokens, $20 per million output tokens, and $0.20 per million cache reads. Cached input is billed at the read price; the examples leave out cache writes.

A Claude Code task is a loop of turns. Cost is set by turn count, cache-read share, output (including thinking), and which model prices those tokens. Holding token counts fixed, Anthropic’s illustrated session falls from $3.50 on Opus 5 to $2.40 on Opus 5.5, about 31% less from the price change alone (2.0 million cache reads, 200K fresh input, 60K output). Cache reads are the largest token line on that receipt, which is why the session moves more than the 20% input-and-output cut. A separate 40% “costs less to run” figure is Anthropic’s estimate for typical token-billed workloads at default settings, combining lower prices with an assumed drop in tokens per task at Opus 5.5’s medium default.

The Claude Code changelog lists Claude Opus 5.5 (`claude-opus-5-5`) as the default Opus model, with a 1M context window and those same $4/$20 per million token list rates and $0.20 per million cache reads.

• ClaudeDevs on X
• What a task costs on Opus 5.5
• Claude Code changelog

Who feels it

On Pro, Max, or Team, Anthropic says the lower Opus 5.5 price is passed through to plan limits, including cached context, so those limits go about 25% further than on Opus 5. That about-25% stretch is the subscription effect. The 60% cache-read reduction is the API list price, the line that shows up on an API-key bill. Five-hour limits rose on Pro, Max, Team, and seat-based Enterprise. Eligible subscribers get an optional limit reset under Settings > Usage on the web or in Claude Desktop; a reset applies across the account, Claude Code included.

Opus 5.5 defaults to medium effort, one level below Opus 5’s high default. A named level is not the same thinking budget on every model. At a given level, Anthropic says Opus 5.5 thinks more per turn than Opus 5, most of all at xhigh and max, so a level chosen for Opus 5 should not be carried over. Start at medium. It always thinks before it replies, and thinking is billed as output. Effort levels are low, medium, high, and xhigh, plus max for one session. Changing effort mid-session keeps the cache on an API key or Claude subscription; on Amazon Bedrock, Google Cloud’s Agent Platform, or a Claude apps gateway it still clears the cached conversation.

Fast mode is stated at twice standard Opus 5.5 price, $8 per million input tokens and $40 per million output, and up to 2.5× faster as Anthropic’s claim. On a Claude subscription it bills to usage credits. The first request after you turn it on pays the fast-mode input price on the whole conversation, uncached, so Anthropic’s post says to enable it at the start of a session.

The changelog also records that Pro and Team Standard defaults moved from Sonnet to Opus, matching Max, Team Premium, and Enterprise. Cache lifetime in Claude Code is one hour on a subscription and five minutes on an API key or cloud provider, or once a subscription draws usage credits. Cache writes list at 1.25× the input price for a five-minute cache and 2× for a one-hour cache. At 120K tokens, Anthropic’s example is about $0.02 to read and about $0.60 for a five-minute write.

How to try it

Anthropic says Opus 5.5 needs Claude Code v2.1.280 or later, so run `claude update` first. Fill the blog calculator from a real Session block via `/usage` or `/cost`. Compare the same backlog task on Opus 5 and Opus 5.5 with `/model`, and note turns, output tokens, and cost. Anthropic says to run three or four backlog tasks before drawing a conclusion. On a Pro, Max, Team, or Enterprise plan, Anthropic says the dollar figure on that screen is a list-price guide. The blog’s typical-session preset (2.2M input, 91% cache, 60K output, 10 tasks a day, 22 working days) shows Opus 5 at $3.50 per session / $770 a month versus Opus 5.5 at $2.40 / $528, a $242 monthly gap from price alone.

claude update   # Opus 5.5 needs v2.1.280 or later
# then in a session:
/model
/usage

Still unclear

Anthropic does not state how much five-hour limits increased, or who counts as eligible for the optional reset beyond “eligible subscribers.” Calculator figures omit cache writes and batch or volume discounts. Token use on a given codebase still has to be measured locally.