How Much Does the Claude API Cost? Price per Million Tokens for Every Claude Model
Key Takeaways
- Claude's API is priced per million tokens, and the current rate card runs from $0.10 input / $0.50 output on Haiku 5.5 to $10 / $50 on Fable 5.1, a hundredfold range.
- The 5.5 generation cut prices: Opus 5.5 is $4 / $20 against Opus 5's $5 / $25, and Haiku 5.5 is a tenth of Haiku 4.5's $1 / $5 for prompts up to 100,000 tokens.
- Output costs five times input on every model, so a long answer weighs five times as much in the bill as a prompt of the same length.
- Caching and batching are the only discounts that change the bill materially: cache reads cost 10% of the input price on most models, 5% on Opus 5.5 and Sonnet 5.5 and 2.5% on Fable 5.1, and the Batch API halves both input and output.
- Subscriptions and the API are billed separately; Max and Team plans now include monthly API credits, Pro does not.
- Anthropic has issued no cryptocurrency, and "Claude token" means a unit of text, not a coin; the company has applied to list its shares on Nasdaq as ANT in an IPO it publicly filed for on October 26, 2026.
The Claude API has no monthly fee and no seat price.
It bills for text, measured in tokens, and input prices run from $0.10 to $10 per million tokens depending on the model, with output prices from $0.50 to $50.
That range is wide enough that choosing a model matters more than negotiating anything, and the discounts for caching and batching matter more than either.
This guide gives you the current rate card, which versions are still sold, how to estimate a job, and why none of this has anything to do with a crypto token.
What Is the Claude API Price per Million Tokens?
Billing is per million tokens, split into input, the text you send, and output, the text the model writes.
The figures below are from Anthropic's published rate card and apply to any account with an API key.
| Model | Input | Output | Cache read | Batch input | Batch output |
|---|---|---|---|---|---|
| Fable 5.1 | $10 | $50 | $0.25 | $5 | $25 |
| Mythos 5.1 (verification required) | $10 | $50 | $0.25 | $5 | $25 |
| Fable 5 | $10 | $50 | $1 | $5 | $25 |
| Mythos 5 (verification required) | $10 | $50 | $1 | $5 | $25 |
| Opus 5.5 | $4 | $20 | $0.20 | $2 | $10 |
| Opus 5 | $5 | $25 | $0.50 | $2.50 | $12.50 |
| Sonnet 5.5 | $2 | $10 | $0.10 | $1 | $5 |
| Sonnet 5 | $2 | $10 | $0.20 | $1 | $5 |
| Haiku 5.5 (prompt ≤100K tokens) | $0.10 | $0.50 | $0.01 | $0.05 | $0.25 |
| Haiku 5.5 (prompt >100K tokens) | $0.50 | $2.50 | $0.05 | $0.25 | $1.25 |
| Haiku 4.5 | $1 | $5 | $0.10 | $0.50 | $2.50 |
Cache writes cost 1.25x the input price for a five-minute cache and 2x for a one-hour cache on every model.
Haiku 5.5 is the only model with length-tiered pricing: a request whose prompt exceeds 100,000 tokens bills at the higher tier.
How to Read a Rate Card: Input, Output, and Cache Columns
MTok means one million tokens.
Input includes your messages, the system prompt, the replayed conversation history and any tool definitions; output covers only the text the model generates, and it costs five times the input rate on every row.
The card changed repeatedly in 2026; the three changes that matter for today's bill are Sonnet 5's introductory price made permanent on August 10, Opus 4.1 retired on August 5, and the 5.5 models arriving between September 22 and October 7 at or below the prices of the models they replaced.
Trust Anthropic's page over any secondary source, including this one, when they disagree.
Which Claude Models Can You Actually Buy Today?
The self-serve rate card carries the current generation, Fable 5.1, Opus 5.5, Sonnet 5.5 and Haiku 5.5, alongside Fable 5, Opus 5, Sonnet 5 and Haiku 4.5, and the older Opus 4.5–4.8, Sonnet 4.6 and, until November 30, 2026, Sonnet 4.5 at their original prices.
Mythos 5.1 and Mythos 5 are priced on the same card but sit behind Anthropic's verification programs rather than self-serve signup.
Mythos access is limited to a small set of vetted organizations.
Entry runs through Anthropic's limited-availability program rather than the Console.
Fable 5.1 is the same underlying model as Mythos 5.1 with additional safeguards, which is why the two share a price, and for most teams Fable 5.1 is the top model available.
Opus vs Sonnet vs Haiku vs Fable: The Hundredfold Price Gap
The cheapest and the most expensive Claude models are a hundred times apart: Haiku 5.5 at $0.10 / $0.50 and Fable 5.1 at $10 / $50 per million tokens.
Between them, Sonnet 5.5 at $2 / $10 and Opus 5.5 at $4 / $20 are where most production work lands.
Four factors decide which tier a job belongs on: the context it needs, the cost of a wrong answer, how much it benefits from multi-step reasoning, and only then the cost per token.
Haiku 5.5 suits high-volume classification, extraction and tagging; Sonnet 5.5 is the sensible production default; Opus 5.5 fits multi-step reasoning where errors compound; Fable 5.1 is positioned for agentic coding and long-horizon problems.
Cost per completed job beats cost per token.
A cheap model that needs three attempts costs more than an expensive one that finishes in one, so pick the cheapest model that completes the task, not the cheapest model.
The cheapest and dearest Claude models are 100 times apart, and output always costs five times input. Source: Claude Developer Platform pricing page (list prices).
Which Claude Version Numbers Are Current, and Which Are Retired?
Several older versions remain callable at their listed prices, while others have been retired on a published schedule.
Search results still show retired prices, which is a common source of budgeting errors.
| Model | Status | Price (input / output) |
|---|---|---|
| Opus 5.5, Sonnet 5.5, Haiku 5.5, Fable 5.1 | Current generation | $4/$20 · $2/$10 · $0.10/$0.50 · $10/$50 |
| Fable 5 | Active | $10 / $50 |
| Opus 5, Opus 4.8, 4.7, 4.6 | Active | $5 / $25 |
| Opus 4.5 | Active; eligible for retirement from November 24, 2026 | $5 / $25 |
| Sonnet 5 | Active (introductory price made permanent August 10, 2026) | $2 / $10 |
| Sonnet 4.6 | Active | $3 / $15 |
| Sonnet 4.5 | Deprecated September 30, 2026; retires November 30, 2026 | $3 / $15 |
| Haiku 4.5 | Active; earliest retirement date has passed | $1 / $5 |
| Opus 4.1 (Aug 5, 2026), Opus 4 and Sonnet 4 (Jun 15, 2026), Sonnet 3.7 and Haiku 3.5 (Feb 19, 2026), Haiku 3 (Apr 20, 2026), Sonnet 3.5 (Oct 28, 2025), Opus 3 (Jan 5, 2026) | Retired on the first-party API | Historic: Opus 4.1 was $15 / $75; Haiku 3.5 was $0.80 / $4 |
The dates come from Anthropic's model deprecations page.
A deprecated model keeps running at its old price until the retirement date; a retired one is gone from the first-party API.
Sonnet 5's $2 / $10 introductory price became permanent on August 10, 2026; the $3 / $15 standard rate due on September 1 "no longer applies," per Anthropic's changelog.
One more line to budget for: Claude 4.7 and later models use a tokenizer that produces approximately 30% more tokens for the same text, so a job costs more at the same per-token rate.
Older Claude models leave the rate card on published dates, and deprecated ones keep their old price until then. Source: Claude Developer Platform model deprecations and pricing pages.
Why Is "Claude Token Price" Not a Cryptocurrency Price?
A Claude token price is the API rate for processing text, not the price of a coin.
A token here is a unit of text, roughly four characters, and Anthropic has issued no cryptocurrency, coin or airdrop.
Tokens using the Claude or Anthropic name do appear on public crypto trackers, and none of them is affiliated with the company.
The company's own path to a public market runs through a stock listing, covered in the guide to whether Anthropic stock is public yet.
What Anthropic Says About Unauthorized Offers
Anthropic's investor warning page states that transfers of its stock the board has not approved are void and are not recognized on its books, and that it does not permit special purpose vehicles to acquire its stock.
It warns that offers made through direct sales, forward contracts, tokenized securities or other mechanisms are likely offering an investment that may have no value because of its transfer restrictions, and that anyone providing stock certificates is very likely engaged in fraud.
The warning signs it lists are the usual ones: unsolicited contact, claims of exclusive access, requests for payment in crypto or by wire, pressure to act quickly, workarounds for transfer restrictions, and no documentation of board approval.
What Does Exist: A Futures Contract, Not a Token
MEXC lists an Anthropic pre-IPO futures contract, a USDT-margined perpetual that conveys no shares, voting rights, dividends or claim on the company.
Anthropic filed its public S-1 on October 26, 2026 and has applied to list on Nasdaq as ANT; until the IPO prices there is no share price, and the contract tracks the expected valuation, converting into standard ANT stock futures once the listing completes.
A cash-settled contract makes no ownership claim.
MEXC documents what changes when a pre-IPO contract converts to standard stock futures after an IPO.
How Do You Calculate What a Claude API Job Will Cost?
The formula is input tokens times the input rate plus output tokens times the output rate, each divided by one million.
One token is roughly four characters or 0.75 English words, so a million tokens is about 750,000 words.
A 13,000-token prompt with a 1,300-token answer costs about $0.078 on Opus 5.5 (13,000 × $4 + 1,300 × $20, divided by a million), about $0.039 on Sonnet 5.5, and under a cent on Haiku 5.5.
Anthropic's own worked example on its pricing page is a one-hour Claude Managed Agents session on Opus 5.5 that consumes 50,000 input and 15,000 output tokens: $0.20 of input, $0.30 of output and $0.08 of session runtime, $0.58 in total, or about $0.43 if 40,000 of the input tokens are cache reads.
Estimates give you a budget, not an invoice.
Anthropic's free token-counting endpoint returns exact, model-specific counts before you send a request.
Multiply each side's tokens by its per-million price and add them up. Source: Claude Developer Platform pricing page (list prices).
Is the Claude Subscription Price the Same as the API Price?
No. Subscriptions buy flat-rate access to the Claude apps, while API keys bill per token with no monthly fee, and the two are separate bills.
On Anthropic's pricing page Claude Pro is $20 a month, Max is $100 or $200 for higher usage limits, and Team and Enterprise are priced per seat, with Enterprise API usage billed separately.
The plan-by-plan prices and what each tier includes are in Claude pricing for every plan.
Since 2026 the two are linked in one direction: Max includes $100 or $200 a month of API credits and Team seats pool $20 or $100 each, capped at $500, while Pro includes none.
The credits work on the API and Managed Agents, not inside Claude Code or the apps, and expire monthly.
A subscription suits steady interactive use.
Per-token billing suits bursty programmatic use, where a flat fee would sit idle most of the month.
Subscriptions and the API are separate bills; Max and Team credits link them in one direction only. Source: Claude pricing page, Claude Help Center and Claude Developer Platform docs.
What Does Claude Code Cost?
Claude Code has no separate price.
It comes included in the paid Claude plans, or it bills per token against an API key at standard model rates.
On an API key the cost depends on how much context each session sends, because every message replays the conversation so far.
The usage command inside Claude Code shows an estimated dollar figure, which is an estimate rather than the bill; the subscription-versus-API math is in the guide to Claude Code's subscription price versus API price.
Do Third-Party Platforms Charge a Different Claude Price?
Third-party channels generally match Anthropic's per-token rates and add their fees elsewhere, in platform charges or on credit purchases.
Cloud platforms charge a 10% premium for regional or multi-region endpoints over global ones.
On the first-party API, pinning inference to the United States adds a 1.1x multiplier on Claude 4.6 and later.
Check the channel's own price list before assuming the headline rate applies.
How Can You Lower Your Claude API Bill?
Prompt caching is the largest lever.
A cache hit costs 10% of the input rate on most models, 5% on Opus 5.5 and Sonnet 5.5, and 2.5% on Fable 5.1 and Mythos 5.1; a five-minute cache write costs 1.25x input and breaks even after one read, while a one-hour write costs 2x and needs two reads.
The Batch API takes 50% off both input and output for work that can wait.
Route simple extraction and classification to Haiku 5.5, and keep its prompts under 100,000 tokens to stay on the lower tier.
Claude 4.6 and later models include the full one-million-token context window at standard pricing, except Haiku 5.5, whose prompts over 100,000 tokens bill at the higher tier, so chunking is no longer needed for cost reasons.
Web search costs $10 per 1,000 searches on top of token charges, and code execution is free when it runs alongside web search or web fetch.
Caching and batching are the discounts that move a Claude API bill. Source: Claude Developer Platform pricing page (list prices).
FAQ
How much does the Claude API cost?
From $0.10 input / $0.50 output per million tokens on Haiku 5.5 to $10 / $50 on Fable 5.1, with Sonnet 5.5 at $2 / $10 and Opus 5.5 at $4 / $20.
What is Claude token pricing per million tokens?
It is the API rate card above, quoted per million tokens of text, and it has nothing to do with a cryptocurrency.
Is the Claude API cheaper than a Pro or Max subscription?
For light or bursty use, usually yes; for steady daily use, a flat plan usually costs less, and Max now includes API credits worth its own price.
How much does Claude Opus 5 cost?
$5 input / $25 output per million tokens; its successor, Opus 5.5, costs $4 / $20.
How much does Claude Fable 5.1 cost per token?
$10 per million input tokens and $50 per million output tokens, with cache reads at $0.25 and batch rates of $5 / $25.
How much does Claude Code cost?
Nothing on top of a paid plan, where it shares the plan's usage pool; on an API key it bills per token at the model's standard rates.
Is there a Claude crypto token?
No; Anthropic has issued no coin or airdrop, and tokens using its name on crypto trackers are unaffiliated with it.
What is the cheapest way to use Claude?
Haiku 5.5 with prompts under 100,000 tokens, cached inputs and the Batch API, which brings the effective price to a few cents per million tokens.
Do Claude Opus 4.7 and 4.8 still work?
Yes, at $5 / $25; the retired models are Opus 4.1, Opus 4, Sonnet 4, Sonnet 3.7, Sonnet 3.5, Haiku 3.5, Haiku 3 and Opus 3.
Conclusion
Rate cards change, and in 2026 they changed downward: Anthropic made Sonnet 5's introductory price permanent in August and launched the 5.5 generation at or below the prices of the models it replaced.
The structure is the stable part: output costs five times input, caching and batching are the discounts that matter, and the cheapest model that completes the job wins.
Verify the figures on Anthropic's site before you commit a budget.
If your interest in the rate card is really an interest in the company, how its valuation and funding rounds stack up is the place to start.
More Claude Pricing Guides
- Did Claude Raise Its Prices? Claude Pricing for Free, Pro, Max, Team and Enterprise
- Are You Overpaying for Claude Code? Claude Code Subscription Price vs API Price
- Is Claude Max Worth $200 a Month? The Claude Max Price, Limits and Who Needs 20x
- Which Is Cheaper, Claude or ChatGPT? Claude vs ChatGPT Price Compared for 2026
- Claude Now Costs INR 2,000 in India: The Real Claude Subscription Price by Country for 2026
- Can Claude AI Predict Crypto and Stock Prices? What It Can and Can't Do
- Is Anthropic Stock Public Yet? Anthropic IPO Date, Ticker, and How to Buy
- Is Anthropic Profitable? Revenue, ARR & Cash Burn Explained
热门新闻
查看更多
Kalshi CPI Futures: Can It Challenge Wall Street?
Kalshi is preparing to enter the institutional inflation derivatives market with proposed exchange-traded futures contracts linked to the U.S. Consumer Price Index (CPI). According to Bloomberg's Octo
DeFi Price Manipulation Attacks Hit 43 in 2026
DeFi price manipulation attacks have reached a new reported annual high in 2026, exposing persistent weaknesses in how decentralized financial protocols value assets, manage collateral, and respond to
相关文章
查看更多
MEXC On-Chain Daily Report: S&P Global Completes OpenZeppelin Acquisition to Expand On-Chain Security Risk Assessment
Updated: October 27, 2026, 09:30 (UTC+8)｜Author: MEXCHeadlinesS&P Global completes acquisition of OpenZeppelinNBA and CFTC establish prediction market regulatory cooperationPolymarket U.S. monthly
MEXC On-Chain Daily Report: Kalshi and Polymarket Crypto Event Contracts Hit Record $2.7 Billion in Weekly Trading Volume
Updated: October 26, 2026, 09:30 (UTC+8) | Author: MEXCHeadlinesKalshi and Polymarket crypto event contracts hit $2.7 billion in weekly volumeKalshi plans to launch U.S. CPI futures in DecemberECB
![[Complete Guide to Crypto Trading] Beginner's Guide: How to Safely Buy Your First Cryptocurrency?](https://mexc-rainbown-activityimages.s3.ap-northeast-1.amazonaws.com/banner/F20261025184140311BdXQmbfitXY9lI.png)
