Claude Opus 5.5 cost
Every token counted.
$4.00 input. $20.00 output. Per million tokens on the standard Claude API. Your task total also depends on cache writes, cache reads, thinking, tools and billing mode.
- Uncached input
- $4.00/ 1M tokens
- Billed output
- $20.00/ 1M tokens
- Cache reads
- $0.20/ 1M tokens
- Cache writes · 5 minutes
- $5.00/ 1M tokens
- Cache writes · 1 hour
- $8.00/ 1M tokens
Standard first-party API rates. Both cache-write lifetimes are included below. This independent reference is maintained by CreateFaceless.
Your task, item by item.
Enter total usage across every request in one task. Start with an illustration, then replace it with your own numbers.
Estimated cost: $0.57 per task; $57.00 per month.
ILLUSTRATIVE SCENARIOS
Aggregate across a multi-turn task, including both cache-write lifetimes.
Searches & other charges
Search and manually entered charges do not receive token discounts or the US-only multiplier.
Paste API usage JSON
Cache writes require the 5-minute / 1-hour breakdown. Thinking already included in output is never added twice. Import applies reported speed and geography, keeps your monthly task count and resets other charges. Batch is detected only when service_tier reports it; otherwise confirm Standard vs Batch yourself. Separate compaction/advisor/fallback billing is rejected instead of silently omitted.
See the formula and billing assumptions
Task cost = Σ (tokens in each of the five categories × that category’s rate ÷ 1,000,000) + searches × $0.01 + other charges. Monthly cost = task cost × tasks per month.
The five token categories are disjoint. Ordinary input excludes cache reads and cache writes. We apply Batch (0.5×) or Fast (2×), then US-only (1.1×), to token rates. Effort changes the number of tokens used; it does not change the published rate.
One model. Three billing modes.
Every rate below is USD per million tokens. US-only adds 10% to any of these token rates.
| Token category | Standard | Batch | Fast |
|---|---|---|---|
| Uncached input | $4.00 | $2.00 | $8.00 |
| Billed output | $20.00 | $10.00 | $40.00 |
| Cache reads | $0.20 | $0.10 | $0.40 |
| Cache writes · 5 minutes | $5.00 | $2.50 | $10.00 |
| Cache writes · 1 hour | $8.00 | $4.00 | $16.00 |
Fast and Batch cannot be combined. No extra long-context premium applies across the supported 1M-token context window. A task with several calls can consume more than 1M tokens in aggregate. Pricing rules
Cache reads are cheap. Writes still count.
A new cache entry costs $5.00 / 1M tokens for five minutes or $8.00 for one hour. A later hit costs $0.20. If the cache expires or misses, a new write may be billed. Opus 5.5 requires at least 512 cacheable prompt tokens to create a cache entry.
Caching behaviorOutput includes invisible work.
Opus 5.5 uses adaptive thinking, which is always on. Control its depth with effort. Thinking and tool-call output can be billed even when the final answer is short. Use the API’s total output count. Adding its thinking breakdown a second time overestimates the bill.
Thinking usageAPI billing or a Claude subscription?
The same model can have a very different buying path. Choose the billing route before comparing numbers.
Claude API
Pay for usageMetered tokens and applicable tool charges. No Pro or Max subscription is needed. The calculator above estimates this route.
Claude Pro
$20 / monthUS web monthly price. Annual billing is $200 upfront, about $16.67 per month. Usage limits apply; API credits are billed separately.
Claude Max 5×
$100 / monthHigher subscription usage than Pro. The “5×” allowance is not a fixed token budget or a guarantee of five times as many Opus replies.
Claude Max 20×
$200 / monthHigher subscription usage with limits. Check the current plan’s model access and limits before subscribing.
US web prices before applicable tax. Other countries, app stores, Team and Enterprise can have different pricing. Official plans · Why API billing is separate
A session’s list-price estimate is not your subscription bill.
How Claude Code is authenticated determines whether usage draws on a subscription or an API account. Use /usage to inspect your plan’s usage bars and optional usage-credit spend. An API list-price estimate helps understand the workload; it does not tell a Pro or Max user how much remains in their subscription allowance. API billing should be reconciled in the Claude Console.
Compare the same usage.
Your calculator quantities, applied to three published rate schedules. This compares prices, not the cost of achieving identical results.
| Model | Standard input / output per 1M | Your task · standard | Your month |
|---|---|---|---|
| Opus 5.5 | $4.00 / $20.00 | $0.57 | $57.00 |
| Opus 5 | $5.00 / $25.00 | $0.7375 | $73.75 |
| Sonnet 5.5 | $2.00 / $10.00 | $0.295 | $29.50 |
Actual task usage, tokenization, retries and output quality can differ by model. These are illustrative equal-quantity comparisons, not measured savings. Model rates
What else can be on the bill?
Tokens are the starting point. Check tools, runtime and the platform issuing your invoice.
Search & fetch
Server-side web search adds $10 per 1,000 searches ($0.01 each), plus tokens from the results. Web fetch has no separate tool fee; fetched content still consumes tokens. Client-side tool schemas and results also use tokens.
Tool pricesExecution & agent runtime
Code execution is free beyond tokens when the request includes web search or fetch tools dated 20260209 or later. Otherwise, execution has a five-minute minimum, 1,550 free hours per organization per month, then $0.05 per container-hour. Preloaded files can trigger time billing even without a tool call. Managed Agents charge $0.08 per running session-hour; idle time is excluded. That replaces container-hour billing and has no Batch discount. Add applicable runtime under “other charges.”
Runtime rules & allowancesYour cloud platform
Bedrock and Google Cloud are partner-operated: verify their global and regional endpoint pricing. Claude Platform on AWS and Claude in Microsoft Foundry use Anthropic rates converted to marketplace consumption units, subject to offer terms. Fast is first-party API only.
Cloud billing routesCredits, contracts & tax
Prepaid credits, negotiated discounts, currency conversion, tax and an aggregator’s markup can change your payable amount. This page shows public USD list prices. Your provider’s invoice is the billing authority.
Current pricingThe questions behind the search.
Short answers to the billing details that most often change a cost estimate.
How much does Claude Opus 5.5 cost?
Standard first-party API pricing is $4.00 per million input tokens and $20.00 per million output tokens. Cache reads and writes have separate rates. Pro and Max are subscription plans with usage limits. Official API pricing
Is Opus 5.5 free, or included with Pro and Max?
Check the current Claude plan page for access and limits. Model access on claude.com does not provide free API credits. Pro and Max subscriptions and API usage are billed separately. Subscription vs API
Can I turn $20 of Pro into an exact token budget?
No reliable fixed conversion is published. Limits depend on the model and workload, including message length and conversation history. This calculator’s dollar result is an API list-price estimate, not a subscription-allowance meter. Pro usage guide
Does a one-million-token context window cost more per token?
Opus 5.5 uses standard per-token prices across its supported 1M context window. Larger requests still cost more because they use more tokens; there is no extra long-context price tier. Context capacity is not the same as aggregate task usage. Long-context pricing
Why is my cache-heavy task more expensive than I expected?
Cache reads are discounted, but creating a cache entry costs more than ordinary input. Count initial and later writes, both lifetimes, misses and all requests. Our calculator includes five-minute and one-hour writes explicitly. Cache accounting
Are thinking tokens charged twice?
No. Thinking is part of billed output. Use total output_tokens from the API usage, including thinking. Do not add a reported thinking breakdown again. Higher effort can increase consumption, but does not itself change the published per-token price. Thinking billing
Does a 20% lower token price guarantee a 20% cheaper task?
No. Opus 5.5’s standard input and output rates are 20% below Opus 5’s, but a task may use different tokens, caching, effort, retries or tools. Run representative tasks and compare complete usage. The equal-quantity table above is a price illustration. Published model rates
Why doesn’t this estimate match my invoice?
Reconcile the billing model, actual mode, routing, all requests, cached-token counters, tool charges, runtime, discounts and tax. JSON import cannot infer cache-write lifetime or Batch mode if it is absent; runtime costs must be entered separately. Additional compaction, advisor or fallback iterations can be billed separately and may use another model. This importer rejects those cases; reconcile every billed iteration. Marketplace or aggregator invoices may follow their own terms. Billing details
Sources, date and scope.
Reviewed . These are manually verified published rates, not a live pricing feed. Recheck the official sources before a purchasing decision.
- Claude Platform pricing Token rates, modes, geography, tools and cloud billing
- Prompt caching Disjoint usage counters and cache lifetimes
- Claude plans Pro, Max and subscription billing options
- API vs paid subscriptions Separate usage and billing
- Claude Code costs Session estimates and usage visibility
- Extended thinking Output usage and billing
Initial reference with both cache-write durations, explicit billing modes, local usage import and itemized estimates.