Claude API Pricing Comparison: Official vs Third-Party Providers

Compare the bill for the same workload

A Claude API price comparison is useful only when the model, token categories, and route are specified. Start with the workload you expect to run: how much input repeats, how much output you need, and whether a particular provider or region is required. A low input rate alone does not tell you the final bill.

This guide uses claude-sonnet-4-6 as a concrete comparison model. It is an example, not a claim that it is the newest or best model for every task. For the full current API Route catalog, use Pricing; for account setup, use the quickstart.

Claude Sonnet 4.6 token and cache prices

Static external rates verified October 17, 2026. USD per one million tokens, standard online inference, before taxes, credit purchase fees, negotiated discounts, or extra tool charges. Anthropic’s row uses its global first-party API rate card. OpenRouter’s row uses the Anthropic endpoint for anthropic/claude-sonnet-4.6; other endpoints can cost more.

The API Route row is generated from the same backend price snapshot as our pricing pages, refreshed 2026-10-17T08:46:36.219Z. It describes the displayed group, not every key or routing option. Check your key’s available group and actual request logs before committing a budget. A missing rate means “check Pricing,” never zero cost.

Sources: Anthropic pricing, OpenRouter endpoint rates, and API Route Sonnet 4.6 pricing. The OpenRouter API reports dollars per token; multiply by 1,000,000 to compare with this table.

How each provider bills you

Anthropic direct: pay the model’s input, output, and caching rates through your Claude Console billing arrangement. A Claude app subscription does not provide an interchangeable pool of API credit. Check your own account’s payment terms and usage limits.

OpenRouter: inference is deducted from credits at the selected provider’s rates. Its fee policy separates inference pricing from credit purchase fees; confirm the checkout quote for your payment method. Do not add a guessed universal fee to every request. BYOK is a separate arrangement.

API Route: API requests consume account credit under the applicable model and route pricing. The public rates come from the backend; key group and routing choices matter. The amount charged by a payment method and the spendable credit received should be checked on the top-up page. This comparison does not promise a fixed discount, identical upstream infrastructure, or a particular rate limit.

Worked example: 1M input and 200K output tokens

Assume no cache, no tools, and a total of 1,000,000 input tokens plus 200,000 output tokens across your requests. These are budget assumptions, not measured usage.

Cost = input_tokens / 1,000,000 * input_rate
     + output_tokens / 1,000,000 * output_rate

Anthropic direct: 1 * $3 + 0.2 * $15 = $6.00
OpenRouter / Anthropic endpoint: $6.00 in inference credits
API Route / displayed group: $0.60

If you need to compare cash outlay, use effective cost = inference credits consumed × (cash paid / spendable credits received). For example, a hypothetical $105 payment yielding $100 of spendable credit makes $6 of inference cost $6.30 in cash. That is arithmetic, not a quote for any provider. Include taxes separately when applicable.

Output-heavy workloads can change the decision: 100,000 input tokens and 1,000,000 output tokens cost $15.30 at the direct rates above. Estimate your output distribution rather than copying another team’s input/output ratio.

Worked example: repeated context changes the comparison

Consider a fixed 20,000-token prefix used by ten requests within one 5-minute cache window. Ignore unique input and output for this isolated prefix calculation. With one write and nine reads at the direct Sonnet 4.6 rates:

No cache: 10 * 20,000 / 1,000,000 * $3 = $0.6000
5m cache: 20,000 / 1,000,000 * ($3.75 + 9 * $0.30) = $0.1290
1h cache: 20,000 / 1,000,000 * ($6.00 + 9 * $0.30) = $0.1740

These are conditional scenarios, not a promised hit rate. Ten separate cold requests with 5-minute writes would cost $0.75 for that prefix. Compare billed cache reads and writes for your route, not just whether the model supports caching. Our prompt caching guide includes a two-request test and a token-based hit-rate calculation.

Recalculate with your observed token mix

Run this in Node.js or a browser console. Replace the counts with usage totals, and replace the rates with your selected route’s prices. On Anthropic-format usage, uncached input, cache writes, and cache reads are separate categories; do not count the same token in two categories.

const usage = { input: 1000000, output: 200000, write5m: 0, write1h: 0, read: 0 };
const rates = { input: 3, output: 15, write5m: 3.75, write1h: 6, read: 0.3 };
const usd = Object.keys(usage).reduce(
  (sum, key) => sum + usage[key] * rates[key] / 1e6, 0
);
console.log(usd.toFixed(4)); // 6.0000

Keep thinking tokens and provider-specific tool charges in your reconciliation when applicable. A local calculator is an estimate; the provider’s settled usage log is the record to compare against.

Which provider should you buy from?

• Choose direct access when your requirements favor a direct Anthropic relationship, specific first-party features, or controls you have verified in its commercial terms.
• Evaluate OpenRouter when provider selection across models is central to your application. Set the endpoint, privacy requirements, and fallback policy before comparing costs.
• Evaluate API Route when its current model/group pricing and available integration fit your workload. Start with a small representative test before funding sustained traffic.

Before deciding, run the same prompts and tool calls through your candidate routes. Record errors, token categories, total charges, and whether required streaming and tool behavior works. Review data handling, regional availability, support, and limits directly with the provider. A successful text response does not establish full feature parity. We do not claim measured performance or reliability advantages in this guide.

Ready to test API Route? Create an account, inspect current Pricing, then follow the Claude Code setup guide. Verify a first request in Usage Logs before choosing a top-up amount.