Anthropic API pricing in this catalog ranges from $1.00 input and $5.00 output for Claude Haiku 4.5 to $10.00 input and $50.00 output for Claude Fable 5, quoted per million tokens. Cached-input reads are lower than fresh input for every listed Claude model.
How much do Claude models cost?
Anthropic charges separately for fresh input, cached input, and generated output. The example workload below uses 1 million fresh input tokens plus 250,000 output tokens. It excludes cache creation, tools, storage, taxes, batch discounts, and special long-context tiers.
| Claude model | Input | Cached input | Output | OpenRouter in / out | Example cost |
|---|---|---|---|---|---|
| Claude Fable 5 | $10.00 | $1.00 | $50.00 | $10.00 / $50.00 | $22.50 |
| Claude Opus 5 | $5.00 | $0.50 | $25.00 | $5.00 / $25.00 | $11.25 |
| Claude Sonnet 5 | $2.00 | $0.20 | $10.00 | $2.00 / $10.00 | $4.50 |
| Claude Opus 4.8 | $5.00 | $0.50 | $25.00 | $5.00 / $25.00 | $11.25 |
| Claude Opus 4.7 | $5.00 | $0.50 | $25.00 | $5.00 / $25.00 | $11.25 |
| Claude Opus 4.6 | $5.00 | $0.50 | $25.00 | $5.00 / $25.00 | $11.25 |
| Claude Sonnet 4.6 | $3.00 | $0.30 | $15.00 | $3.00 / $15.00 | $6.75 |
| Claude Haiku 4.5 | $1.00 | $0.10 | $5.00 | $1.00 / $5.00 | $2.25 |
Official source: Anthropic pricing documentation. Claude Sonnet 5 currently carries a time-limited introductory-price note in the catalog; always review the source before making a long-term commitment.
How does Anthropic token pricing work?
Fresh input covers the prompt, system instructions, conversation history, and other context processed for a request. Output covers the tokens generated in the response. Cached input can reduce the cost of repeatedly sending the same stable prefix, such as a long system prompt, tool definitions, or reference documents.
Do not calculate a Claude budget from context-window size alone. A large context limit tells you what can fit in a request, not what every request costs. Actual spend depends on tokens processed, the share served from cache, the output length, the number of runs, and any provider-specific tier rules.
Which Claude price tier fits your workload?
Haiku tier
Start here for classification, extraction, routing, and high-volume tasks when the lower-priced model meets your accuracy needs.
Sonnet tier
Use a mid-tier model when you need stronger generation or coding performance but cannot justify the highest output rate for every request.
Opus and Fable tiers
Reserve premium models for workloads where better results are worth a materially higher output-token price.
Should you call Anthropic directly or through OpenRouter?
Direct Anthropic access is the simpler reference point for first-party features, limits, support, and contractual terms. OpenRouter offers a unified API and routing layer across providers. The mapped Claude models currently have the same base token rates in both columns, but OpenRouter account funding fees remain separate.
If the token prices are equal, choose based on integration needs, fallback strategy, data policy, rate limits, and support rather than treating the two routes as identical products.
Anthropic API pricing questions
How much does the Anthropic API cost?
In this catalog, standard Anthropic input prices range from $1.00 to $10.00 per million tokens, while output prices range from $5.00 to $50.00.
Which Claude model is cheapest?
Claude Haiku 4.5 is the lowest-priced Anthropic model in this catalog at $1.00 per million input tokens and $5.00 per million output tokens.
Are cached Claude tokens cheaper?
Yes. The listed Anthropic models have a lower cached-input read price than fresh input. Cache creation and special long-context rules may be billed differently, so check Anthropic's current documentation for the exact workload.
Is Claude cheaper through OpenRouter?
The mapped Claude models currently show the same base inference prices on OpenRouter as the first-party list rates in this dataset. OpenRouter credit purchase fees are separate, so effective account cost can still differ.
Does Anthropic charge for output and reasoning tokens?
Output token charges cover tokens generated by the model. Any model-specific reasoning or tool-related accounting should be checked in the current Anthropic documentation because not every feature uses the same billing treatment.