Claude API Pricing 2026: Every Model's Price per 1M Tokens (Opus 5, Fable 5, Sonnet 5, Haiku 4.5)

Short answer: Anthropic bills the Claude API per token. In 2026 the list prices per 1M tokens are Claude Opus 5 $5 in /$25 out, Claude Sonnet 5 $2 /$10, Claude Haiku 4.5 $1 /$5, and the new Claude Fable 5 $10 /$50. Cached input costs 10% of the input price. The same models are available through Crazyrouter at 35% below list, billed per token with no subscription.

Last updated: September 30, 2026. Prices are re-verified daily against Anthropic's price list; each model links to a page with Azure, Bedrock, Vertex and OpenRouter prices and a monthly cost calculator.

Claude API price table (per 1M tokens, USD)#

Prices are per 1M tokens: 1,000 tokens of input on Opus 5 is $0.005, and a typical 800-token answer is$0.02. Batch (asynchronous) requests are 50% off list on Anthropic's own API. Long-context requests above 200K tokens on the 1M-context models are billed at a higher rate on Anthropic — see the individual model pages for the exact tiers.

Which Claude model should you pay for?#

• Claude Sonnet 5 is the default for coding agents and production apps: $2 /$10 is a fifth of Opus with most of the capability.
• Claude Opus 5 for long agentic runs, hard reasoning and review passes where a wrong answer costs more than the tokens.
• Claude Haiku 4.5 for classification, extraction, routing and anything you call thousands of times.
• Claude Fable 5 is the top tier ($10 /$50) for tasks where Opus is not enough; treat it as a per-task tool, not a default.

What a month actually costs#

Cached prompts change this a lot: a coding agent that re-sends a 30K-token system prompt every turn pays 10% of the input price on the cached part, which is why agent bills are dominated by output tokens rather than input.

How Claude API billing works#

Anthropic sells prepaid credits: you add money in the Console, usage is deducted per request, and monthly spend limits rise through usage tiers. There is no monthly fee and no free tier; see the Anthropic billing guide for credits, auto-reload, invoices and declined-card fixes. If you are wondering whether any free access exists, see Claude API free tier — short version: none, from anyone.

Through Crazyrouter the billing model is the same (prepaid, per token) but the price is 35% lower on every Claude model, you can pay by card, Alipay or WeChat Pay, and one key also covers GPT, Gemini, DeepSeek and the image / video models. The API is Anthropic-compatible, so Claude Code, Cursor and the official SDKs work by changing the base URL — see the Claude Code integration guide.

Frequently asked questions#

How much does the Claude API cost per 1,000 tokens? Divide the per-1M price by 1,000: Sonnet 5 is $0.002 per 1K input tokens and$0.01 per 1K output tokens; Opus 5 is $0.005 /$0.025; Haiku 4.5 is $0.001 /$0.005.

Is there a free Claude API tier? No. Anthropic's API is prepaid only; the free plan on claude.ai is the chat app, not the API. Crazyrouter has no free tier either; it is prepaid pay-as-you-go at 35% below list, so a $5 top-up covers roughly 3.8M Sonnet 5 input tokens of testing — details in the free tier guide.

Which Claude model is cheapest? Claude Haiku 4.5 at $1 /$5 per 1M tokens on Anthropic ($0.65 /$3.25 on Crazyrouter).

Does Claude API pricing differ on Bedrock, Vertex or Azure? The per-token price is the same as Anthropic's list price on all three clouds; you pay for regional residency and consolidated billing, not for a discount. Each model page above shows the cloud prices side by side.

How is Claude Max different from the API? Max is a $100 or$200 monthly subscription to the Claude app and Claude Code with usage caps; the API is metered per token with no cap. The Claude Max guide works through when each is cheaper.

→ Get an API key — pay as you go · All 35 model prices