Home → Guides → Claude Opus 5.5 pricing, and what actually changed
Anthropic priced Opus 5.5 below Opus 5 on every line. The token rates are 20% lower; the cache read rate is 60% lower, and that is the one that moves an agent bill.
From Anthropic's pricing page, per million tokens:
| Opus 5.5 | Opus 5 | |
|---|---|---|
| Input | $4.00 | $5.00 |
| Output | $20.00 | $25.00 |
| Cache write (5m) | $5.00 | $6.25 |
| Cache write (1h) | $8.00 | $10.00 |
| Cache read | $0.20 | $0.50 |
| Batch | $2.00 / $10.00 | $2.50 / $12.50 |
Note the cache read line. Every other Claude model charges a tenth of the input price for a cache hit. Opus 5.5 charges a twentieth — $0.20 against the $0.40 a 0.1× multiplier would give. Only Fable 5.1 goes lower, at 0.025×.
Agent loops re-read the same prefix on every turn: the system prompt, the tool definitions, the document or repository under discussion. Those tokens are cache reads, and in a long session they dominate the token count.
A loop that reads a 100,000-token cached prefix, adds 5,000 fresh input tokens and returns 2,000 output tokens, priced per call:
That is 36% less per call — more than the 20% the headline rates suggest, because the cache portion fell furthest. Over a thousand turns the difference is $45.
The usual caveat applies: a cache write still costs more than fresh input ($5.00 against $4.00). If the front of your prompt changes every turn, none of this discount reaches you. Keep volatile content at the end.
Fast mode, in research preview, is priced at $8.00 / $40.00 on Opus 5.5 against $10.00 / $50.00 on Opus 5 — the same 20% gap, applied across the full context window. The Batch API halves the standard rates on both.
Opus 5.5 is served from the Claude group carrying the higher upstream rate, so the price here is a fixed fraction of Anthropic's list rather than the deepest discount on the platform. The exact figure, recalculated hourly from live billing data, is on the model page; the comparison that matters is against Opus 5 at the same multiplier, where the 20% lower list price carries straight through.
The cache read rate carries through as well, and that is where the money is. A loop that re-reads a 100,000-token prefix fifty times pays for five million cached tokens. At a tenth of the input price that is one figure; at a twentieth it is half of it. Nothing about your code changes — the discount is in the rate card, not in how you call the model.
One caveat worth repeating: a cache write still costs more than fresh input. If the front of your prompt changes on every turn, none of this reaches you.
Opus 5.5 is available here as claude-opus-5-5. What it costs on this
platform, beside the official rate, is on its
pricing page — as are
claude-opus-5 and
claude-sonnet-5 for comparison. The authoritative list for your own
key is always GET /v1/models; if a model is missing there,
this page explains why.
No subscription, no monthly minimum, no sales call. Top up with USDT and spend what you use — 1 USDT gives you 2 credits of API balance.
Read the 30-second quickstartPrices quoted on this page are regenerated automatically from live billing data. Third-party terms are quoted from that party's own published documentation.