Anthropic released Claude Opus 5.5 on September 23, 2026, the first model in its Claude 5.5 family. The company says it performs at the level of Claude Fable 5.1 on most work and costs 40% less to run than Opus 5 on typical workloads.
The per-token list price fell too, but not by 40% across the board. Input tokens are $4 per million, down from $5. Output tokens are $20 per million, down from $25. Cache writes are $5 per million, down from $6.25. All three are 20% cuts. Cache reads are $0.20 per million, down from $0.50, a 60% cut. Anthropic says cache reads make up the majority of agentic and coding work costs.
Where the 40% figure comes from
None of the four listed price cuts is 40% on its own. A workload made entirely of fresh input and output tokens gets a 20% discount from the price table alone. A workload made entirely of cache reads gets 60%. Anthropic’s “40% less… on typical workloads” line sits between those two, which is consistent with a mix of cache reads and fresh tokens, weighted toward cache reads.
That number also is not only a price change. Anthropic separately says Opus 5.5 uses fewer tokens per task: it cut load times across every page of a test web app in 39 of 40 tries where Opus 5 made smaller, behavior-altering fixes, and GitHub’s Chief Product Officer said Opus 5.5 used among the fewest tokens and steps of any model tested in its GitHub Copilot CLI and VS Code testing, solving more terminal tasks than Opus 5 in less than half the steps. Anthropic’s announcement does not say how much of the 40% comes from the lower list price and how much comes from using fewer tokens to do the same work. A buyer who applies the price table to last month’s own token counts, rather than assuming a 40% cut, gets the number that actually applies to their account.
Fast mode moves the other direction. It costs $8 per million input tokens and $40 per million output tokens, more than even Opus 5’s old standard price, in exchange for output more than 30% faster and up to 2.5x speed. Turning on fast mode gives up the discount entirely.
| Token type | Opus 5.5 | Opus 5 | Change |
|---|---|---|---|
| Input | $4/M | $5/M | -20% |
| Output | $20/M | $25/M | -20% |
| Cache write | $5/M | $6.25/M | -20% |
| Cache read | $0.20/M | $0.50/M | -60% |
| Fast mode input | $8/M | n/a | n/a |
| Fast mode output | $40/M | n/a | n/a |
Safeguards can move work onto a different model
Opus 5.5 ships with cybersecurity and biology safeguards similar to Fable 5.1’s. Most cybersecurity tasks route to Claude Opus 4.8 instead when those safeguards intervene, and some biology and frontier-model-development tasks route to Opus 5. Anthropic’s own benchmark footnotes describe the effect on scores: its AutomationBench run, done by Zapier without fallback models, counted every safeguard intervention as a failure, which Anthropic says produced a lower score than Opus 5.5 achieves in practice with fallbacks running.
The same rerouting applies to cost, and the announcement does not price it. A task that reroutes to Opus 4.8 or Opus 5 bills and performs as that model, not as Opus 5.5, so a workload with a meaningful share of cybersecurity or biology work will not see the full discount on that share. Anthropic did not publish how often the fallback triggers outside of the benchmark footnotes, so the size of that share is something a buyer has to measure against their own traffic, not estimate from the announcement.
