As of November 10, 2026, Anthropic lists Claude API prices in USD per million tokens (MTok), with input and output billed separately. The current models shown below range from Haiku 5.5 at $0.10/$0.50 for prompts up to 100,000 tokens to Fable 5.2 at $8/$40. Haiku 5.5 has a higher rate card for longer prompts. (Anthropic pricing)
Claude API prices per million tokens
These are Anthropic’s listed rates for the Claude API—not a guarantee that other cloud-provider channels use the same prices. Check Anthropic’s current price sheet before setting a production budget; rates and model availability can change. (Anthropic pricing)
| Model | Input / MTok | Output / MTok | Cache read / MTok |
|---|---|---|---|
| Claude Haiku 5.5, prompt up to 100,000 tokens | $0.10 | $0.50 | $0.01 |
| Claude Haiku 5.5, prompt over 100,000 tokens | $0.50 | $2.50 | $0.05 |
| Claude Sonnet 5.5 | $2 | $10 | $0.10 |
| Claude Opus 5.5 | $4 | $20 | $0.20 |
| Claude Fable 5.2 | $8 | $40 | $0.20 |
Haiku’s rate tier is determined per request. Anthropic says prompt length includes all input tokens, including cache reads and cache writes. Use the rate card for each request’s prompt length, then total the costs across requests; don’t apply one tier to a whole month that contains requests on both sides of the threshold. (Anthropic pricing)
Estimate your API bill
For a standard Messages API estimate, calculate each usage category at its applicable rate, then add the charges:
Token cost = (uncached input × input rate + cache-write input × write rate + cache-read input × read rate + billed output × output rate) ÷ 1,000,000.
Use the API response’s usage counts rather than estimating from visible text alone. With prompt caching, input usage is divided among uncached input, cache-creation input, and cache-read input; count each category at its own rate instead of charging cached tokens again at the base input rate. Internal thinking tokens are billed as output tokens, even when they are not visible in the response—do not add the thinking-token breakdown a second time. (Prompt caching documentation, thinking documentation)
For example, 1 million input tokens plus 250,000 output tokens on Sonnet 5.5 costs $4.50 at standard rates: (1 × $2) + (0.25 × $10). This example assumes no caching or Batch discount and excludes separately billed services or other applicable modifiers. For a monthly estimate, use measured monthly usage or sum request-level estimates, keeping input and output separate.
Prompt caching
Anthropic lists five-minute cache writes at 1.25× the base input rate and one-hour writes at 2×. Cache-read rates vary by model and, for Haiku 5.5, by prompt-length tier. For example, Sonnet 5.5 reads cost $0.10/MTok; its five-minute and one-hour writes cost $2.50 and $4/MTok. A discounted read lowers the cost only when a request actually reads from cache, so check usage data before relying on projected savings. (Anthropic pricing)
Batch API
For eligible asynchronous work, Anthropic says the Message Batches API reduces costs by 50%. At that discount, Sonnet 5.5 is $1/MTok input and $5/MTok output; Opus 5.5 is $2/$10; Fable 5.2 is $4/$20. Batch can be combined with prompt-caching discounts. It is intended for work that does not require an immediate response. (Anthropic pricing)
Choosing a model for a production workload
Use the price table as a starting point, not as proof that the lowest per-token rate will produce the lowest cost per successful task. Test representative requests on the models you are considering. Compare output quality against your requirements, measured input and output usage, and the resulting bill. A lower-priced option may be sufficient for routine tasks; choose a more expensive option when your evaluation shows its results justify the additional cost.
Are Claude API costs included in a Claude subscription?
Eligible Max, Ultra, and Team plans include monthly credits for Claude API use. The reported allowances are $100 for Max, $200 for Ultra, and Team credits of $20 per Standard seat or $100 per Premium seat, pooled up to $500. Free, Pro, and Enterprise plans are not eligible. (Monthly API credits support page)
Check Console for your eligibility and credit-claiming steps, and confirm which usage qualifies before budgeting around the credits. Unused credits expire at the end of the billing cycle and do not roll over. (Monthly API credits documentation)
The estimate above is the gross usage cost; eligible monthly credits may offset covered API charges. For prepaid accounts, if credits run out, API calls and playground use stop until more credits are added. Check your Console billing and payment arrangement for how charges are handled in your account. (API usage payments help page)
There is no universal API-versus-subscription break-even: compare your measured API usage, any credits you can claim, subscription fees, and the applicable plan limits.
Official pricing and billing references
For current rates and conditions, consult Anthropic’s pricing documentation and its pages on prompt caching, thinking, monthly API credits, and API usage payments.
Sources
Pricing — Claude Platform Docs
Prompt caching — Claude Platform Docs
Thinking — Claude Platform Docs
API credits for Max, Ultra, and Team plans — Claude Platform Docs
Monthly API credits for Max, Ultra, and Team plans — Claude Help Center