Claude Opus API pricing at a glance
Anthropic’s pricing documentation lists Claude Opus 5.5 at $4 per million input tokens and $20 per million output tokens for the Claude API. The model ID is claude-opus-5-5; prices are in USD. The rate card lists separate prices for cache writes and reads, and a discount for eligible Batch API requests. (Anthropic pricing · Opus 5.5 model overview)
| Claude Opus 5.5 API usage | Price per 1 million tokens |
|---|---|
| Standard input | $4.00 |
| Standard output | $20.00 |
| 5-minute prompt-cache write | $5.00 |
| 1-hour prompt-cache write | $8.00 |
| Prompt-cache read or refresh | $0.20 |
| Message Batches API input and output | 50% off standard rates |
How to apply cache and Batch rates
Prompt caching applies to eligible prompt prefixes, not automatically to every input token. For Opus 5.5, the model documentation gives a minimum cacheable prompt length of 512 tokens. Caching must be enabled with a cache breakpoint; a later request can use the cached prefix when it matches. The default cache lifetime is five minutes, and Anthropic also offers a one-hour option. (Opus 5.5 model overview · Prompt caching documentation)
For cost accounting, use the response’s separate input_tokens, cache_creation_input_tokens, and cache_read_input_tokens usage fields with their corresponding standard-input, cache-write, and cache-read rates. Anthropic defines total input tokens as the sum of those three fields, so do not count the same usage again under standard input after applying a cache rate. The pricing table lists the rates for each category separately. (Prompt caching documentation · Anthropic pricing)
The 50% Batch discount is for requests submitted through Anthropic’s Message Batches API, which processes them asynchronously. It suits work that can wait for results rather than requiring an immediate response. Standard synchronous requests do not qualify just because they are part of a large workload. Anthropic’s documentation says prompt-caching and Batch discounts can be combined. (Batch processing documentation)
Estimate your production API bill
For standard, uncached, non-Batch usage, estimate the token subtotal as:
(monthly input tokens ÷ 1,000,000 × $4) + (monthly output tokens ÷ 1,000,000 × $20)
For example, a hypothetical workload of 10,000 requests per month, averaging 3,000 input tokens and 1,000 output tokens per request, would use:
- Input: 10,000 × 3,000 = 30 million tokens; 30 × $4 = $120.
- Output: 10,000 × 1,000 = 10 million tokens; 10 × $20 = $200.
- Estimated standard-rate subtotal: $320 per month.
This is an illustration, not a measured workload or quote. It assumes all usage is billed at standard rates and excludes cache pricing, Batch API, Fast mode, US-only inference, and retries or tool-driven follow-ups. Input includes the full prompt—such as system instructions, conversation history, tool definitions, and tool results—not just the latest user message. Anthropic also bills Opus 5.5 thinking tokens as output tokens, even when the thinking is not shown in the response. (Token-counting documentation · Thinking documentation)
For a production budget, replace the example averages with measured usage from representative requests. Track input and output separately, and account for cached and uncached tokens in their respective categories.
Other pricing details that can change the estimate
- US-only inference: Anthropic states that US-only inference is available at 1.1× pricing for input and output tokens. Applied to Opus 5.5’s base rates, that is $4.40 per million input tokens and $22 per million output tokens. This statement is limited here to those base input and output rates; do not assume the multiplier applies to cache rates without checking the applicable rate card. (Anthropic’s Opus page)
- Fast mode: Anthropic lists Opus 5.5 Fast mode at $8 per million input tokens and $40 per million output tokens—a separate rate, not the standard price. (Anthropic’s Opus page)
- Older Opus versions: Anthropic’s pricing table lists Opus 5 and Opus 4.5, 4.6, 4.7, and 4.8 at $5 per million input tokens and $25 per million output tokens. Compare measured cost and task results for your own workload before choosing a version. (Anthropic pricing)
- Model fit: Anthropic lists Opus 5.5 with a 1-million-token context window, 128,000-token maximum output, and text-and-image input. Confirm that the model and limits suit your workload and API channel. (Opus 5.5 model overview)
- Cloud-provider billing: The prices above are Anthropic’s Claude API rates. If you deploy through a cloud provider, check that provider’s rate card for the specific channel and region.
Keep API charges separate from Claude subscriptions. A Pro, Max, or Ultra plan fee—and any plan-linked API credit—is not the per-token price for a production Opus API workload.
Sources
Pricing — Claude Platform Docs
Claude Opus 5.5 — Claude Platform Docs
Prompt caching — Claude Platform Docs
Batch processing — Claude Platform Docs