This Claude API cost calculator estimates first-party Anthropic token charges for active Claude Fable, Opus, Sonnet and Haiku models. Enter the separate uncached-input, prompt-cache and output usage from a representative request, then multiply it by your request volume.
Claude pricing changes with processing mode, cache operation and—on supported models—inference geography. Web search also has a per-search fee. Confirm material budgets against Anthropic’s official Claude API pricing.
Estimate your Claude API cost
Pricing last checked: August 19, 2026. All estimates are in USD.
Claude Sonnet 5 · Standard · Global · 1,000 requests
- Uncached input
- $2.00
- 5-minute cache writes
- $0.00
- 1-hour cache writes
- $0.00
- Cache reads
- $0.00
- Output
- $5.00
- Web search
- $0.00
- Token subtotal
- $7.00
- Average per request
- $0.007
Planning estimate only. Actual usage, retries, tool output, cache hits and contract terms can change the billed amount.
Current Claude API rates used by this calculator
All figures below are USD per million tokens for first-party Claude API use. Batch cache rates are also discounted by 50% and are applied automatically above.
| Models sharing the rate | Input | 5m write | 1h write | Cache read | Output | Batch input | Batch output |
|---|---|---|---|---|---|---|---|
| Fable 5; Mythos 5 limited access | $10 | $12.50 | $20 | $1 | $50 | $5 | $25 |
| Opus 5, 4.8, 4.7, 4.6, 4.5 | $5 | $6.25 | $10 | $0.50 | $25 | $2.50 | $12.50 |
| Sonnet 5 | $2 | $2.50 | $4 | $0.20 | $10 | $1 | $5 |
| Sonnet 4.6, 4.5 | $3 | $3.75 | $6 | $0.30 | $15 | $1.50 | $7.50 |
| Haiku 4.5 | $1 | $1.25 | $2 | $0.10 | $5 | $0.50 | $2.50 |
Fast uses $10 per million input tokens and $50 per million output tokens for supported Opus models. US-only inference multiplies each token rate by 1.1. The $0.01 successful web-search fee is not multiplied.
How to estimate your Claude API bill
- Choose the exact model and processing mode. Batch can reduce token charges when asynchronous processing suits the workload. Fast is limited to supported Opus models and may require approved access.
- Copy token counts from a representative API response. Keep uncached input, five-minute writes, one-hour writes, cache reads and output separate because Anthropic prices them differently.
- Enter the request volume. For a monthly forecast, use the number of calls expected in a typical month. Decimal web searches are accepted as an average per request.
- Compare the result with observed usage. Re-run the estimate after prompt, model or caching changes, then reconcile it with Anthropic’s usage and cost reports.
How Claude prompt caching changes cost
Claude reports uncached input, cache creation and cache reads as separate usage categories. A five-minute cache write costs 1.25 times the model’s base input rate, a one-hour write costs 2 times the base input rate, and a cache read costs 0.1 times the base input rate before other applicable modifiers.
The cache fields above are not subsets of uncached input. Anthropic defines total input as uncached input plus cache-creation input plus cache-read input. For the best forecast, copy the relevant values from the response usage object rather than estimating all prompt text as one input number.
Standard vs Batch vs Fast mode
- Standard uses the model’s normal token rates and suits synchronous requests.
- Batch applies the Message Batches API’s 50% token discount. Prompt-cache multipliers stack with Batch, while a successful web search still costs $0.01.
- Fast is an Anthropic research preview for Claude Opus 5 and Opus 4.8. It uses $10 per million input tokens and $50 per million output tokens across the full context window. Fast cannot be combined with Batch.
Does Claude charge more for long context?
Claude 4.6 and later models include their one-million-token context window at the normal per-token rate; this calculator does not add a long-context uplift. A larger prompt still costs more because more tokens are billed.
Claude 4.7 and later use a newer tokenizer that Anthropic says may produce about 30% more tokens for the same text, depending on the workload. That is not a pricing multiplier. Use actual API usage when comparing old and new models.
Web search and other Claude tool charges
Anthropic lists web search at $10 per 1,000 successful searches, plus normal model tokens for search-generated content. Searches that return an error are not billed. Web search inside a Batch request has the same per-search charge.
Web fetch has no separate per-call fee, although fetched content still produces token charges. Code execution has organization-level free allowances and container-hour pricing that do not map cleanly to one request, so it is not included here. Client tools, bash, text editing and computer use generally add token or tool-definition overhead rather than one stable per-call fee.
What this estimate does not include
This calculator covers first-party Claude API token usage and successful web-search calls. It does not estimate claude.ai subscriptions, Amazon Bedrock, Google Cloud, Microsoft Foundry, marketplace billing, Priority Tier, Claude Managed Agents, negotiated discounts, taxes, retries hidden by your application or surrounding infrastructure.
Claude API calculator FAQ
Is the Claude API included with a Claude Pro or Max subscription?
No. Consumer Claude subscriptions and first-party Claude API usage are separate products. This page estimates API usage, not subscription allowances.
How much does Claude Sonnet 5 cost?
When checked on August 19, 2026, Anthropic listed Claude Sonnet 5 at $2 per million input tokens and $10 per million output tokens for standard global API usage. Anthropic’s current pricing page says these are now the standard rates and the previously planned September increase will not occur.
Does Batch reduce Claude prompt-cache pricing?
Yes. Anthropic says the 50% Message Batches discount and prompt-caching multipliers stack. Cache hits are best-effort in asynchronous batches, so estimate from observed usage where possible.
Does a one-million-token prompt cost extra?
Claude 4.6 and later include their full one-million-token context at the normal per-token rate. There is no separate long-context surcharge, but the larger number of billed tokens still raises the total cost.
Why are cache writes and cache reads separate?
They have different rates. A five-minute cache write costs 1.25 times base input, a one-hour write costs 2 times base input, and a cache read costs 0.1 times base input before other applicable modifiers.
How much does Claude web search cost?
Anthropic lists web search at $10 per 1,000 successful searches, plus the model tokens consumed by search-generated content. Failed search calls are not billed.
Why can the estimate differ from my invoice?
Actual prompts, outputs, cache hits, retries, tool use and enterprise terms vary. Claude 4.7 and later also use a newer tokenizer that may produce more tokens for the same text than earlier models. Reconcile forecasts with the API usage and cost reports.
Sources and related calculators
This is an independent calculator, not an Anthropic product. Rates and compatibility were checked against Anthropic’s pricing documentation, models overview, prompt-caching guide, Batch guide, Fast-mode guide, data-residency guide and web-search documentation.
Compare providers with our AI API cost calculators, OpenAI API cost calculator and Gemini API cost calculator.