Claude's API pricing is easy to get lost in: many models, rates by the token, and a bill separate from your plan. We'll show you what your project will cost, how to cut it, and when a subscription pays off.
๐ฏ Quick Answer: How much does the Claude API cost?
Claude API pricing comes down to one decision: which model you call. Big bills rarely come from the question you ask. They come from the output and the context you resend on every call. A Claude Pro subscription covers none of it: the API is billed separately, per token.
Per million input tokens, prices for current models run from $0.10 on Haiku 5.5 to $8 on Fable 5.2, and each step up, from Sonnet to Opus to Fable, doubles the price.
A support bot answering 10,000 questions a month (about 3,000 tokens in and 400 out each, thinking included) costs about $5 on Haiku 5.5, $100 on Sonnet 5.5 and $200 on Opus 5.5.
Claude API token pricing for Haiku, Sonnet, Opus and Fable
US dollars per million tokens, from Anthropic's pricing page. The Batch API halves every price.
| Model | Best for | Input / output | Cached input |
|---|---|---|---|
| Haiku 5.5 | High-volume classification, extraction and routing | $0.10 / $0.50 | $0.01 |
| Sonnet 5.5 | Everyday coding, agent and production workloads; fast and capable | $2 / $10 | $0.10 |
| Opus 5.5 | Long-running agentic coding and knowledge work | $4 / $20 | $0.20 |
| Fable 5.2 | The most demanding reasoning and long agent runs | $8 / $40 | $0.20 |
What changes these prices:
- Long prompts on Haiku 5.5: a prompt over 100,000 tokens pays five times every rate, cache and batch included ($0.50 in, $2.50 out). The other current models charge one rate across their full 1-million-token context.
- Cache writes: storing a prompt in the cache costs 1.25 times the input price for 5 minutes, or twice the input price for an hour.
- Fast mode: where offered, up to 2.5 times faster output at double the price. Opus 5.5 goes to $8 in and $40 out, Fable 5.2 to $16 and $80.
- US-only processing: adds 10% to every rate on Claude 4.6 and later models.
- Web search: $10 per 1,000 searches.
What makes a Claude API bill bigger
- Output. Every Claude model charges five times more for output than for input, so long answers cost more than long questions.
- Thinking. Claude 5 models think before answering by default, and thinking is billed as output even when you don't see it.
- Conversation history. Chats and agents resend the conversation so far with each request, so input grows every turn.
- Tools. Tool definitions add input tokens to each request that uses them.
Does a Claude subscription include API access?
Only Max, Ultra and Team do, as a monthly credit. Free, Pro and Enterprise include no API usage. The API runs in a separate Claude Console organization with its own balance, and API usage is never charged to your Claude plan.
| Plan | Monthly API credits | Plan price |
|---|---|---|
| Free | None | $0 |
| Pro | None | $20/month, or $17/month billed annually |
| Max | $100 | $100/month |
| Ultra | $200 | $200/month |
| Team | $20 per Standard seat, $100 per Premium seat, pooled up to $500 | $25 or $125 per seat monthly ($20 or $100 billed annually) |
| Enterprise | None; plan usage is billed at API rates | $20 per seat/month billed annually, 20-seat minimum |
- Claiming them: after 7 days on the plan, link a Console organization in your claude.ai billing settings. No payment method needed.
- What they cover: the Claude API, Managed Agents, the Agent SDK and the Console playground, but not extra usage in the Claude apps.
- When they end: credits expire each billing cycle with no rollover. If they run out and you haven't bought credits, API requests stop until next month's arrive.
$100 of credits covers 200,000 of the support-bot replies above on Haiku 5.5, 10,000 on Sonnet 5.5 (that bot's whole month) or 5,000 on Opus 5.5.
Still, a plan pays off only if you'd use the Claude apps or Claude Code anyway. If you only want to build with the API, open a Console account and pay per token.
If you were after the plans themselves, Claude pricing compares every plan and its limits.
What older Claude models cost (Haiku 4.5, Opus 4.6, Sonnet 4.6, Fable 5)
Older models stay on the price list, and most cost more for the same work than their newest replacement.
| Model | Input / output | Status | Cheaper move |
|---|---|---|---|
| Fable 5.1, Fable 5 | $10 / $50 | Active | Fable 5.2: 20% less |
| Opus 5, 4.8, 4.7 | $5 / $25 | Active | Opus 5.5: 20% less |
| Opus 4.6, 4.5 | $5 / $25 | Active; Opus 4.5 guaranteed only to Nov 24, 2026 | Opus 5.5: about the same |
| Opus 4.1, Opus 4 | $15 / $75 | Retired on the Claude API | Opus 5.5: about 65% less |
| Sonnet 5 | $2 / $10 | Active | Sonnet 5.5: same price, half-price cache reads |
| Sonnet 4.6 | $3 / $15 | Active | Sonnet 5.5: about 13% less |
| Sonnet 4.5 | $3 / $15 | Deprecated; retires Nov 30, 2026 | Sonnet 5.5: about 13% less |
| Haiku 4.5 | $1 / $5 | Active | Haiku 5.5: about 87% less |
Why โaboutโ? Claude Opus 4.7 and later models turn the same text into roughly 30% more tokens, so moving to a 5.5 model from anything older (Haiku 4.5, Sonnet 4.6 and earlier, Opus 4.6 and earlier) gives back part of the lower rate. The figures above allow for that. Opus 4.7 to Opus 5.5, both on the newer tokenizer, gets the full cut.
If you see Opus quoted at $15/$75, that's Opus 4.1 or Opus 4, which no longer run on the Claude API.
Two more changes to know about. Sonnet 5's $2/$10 launch price is now permanent; the planned rise to $3/$15 was cancelled. And Anthropic's commitment to keep Haiku 4.5 available ran only to October 15, 2026, so it can now be retired on 60 days' notice.
How to estimate your Claude API cost per month
Multiply your monthly input tokens by the input rate and your output tokens by the output rate, then divide each by a million. The token counts are the hard part, so measure them:
- Count a typical request's input with the free token counting endpoint, using the exact model you'll run.
- Send a few real requests and read the token counts in each response's usage field, not the visible text: billed output includes thinking you don't see.
- Multiply the per-request tokens by your requests per month, then apply the rates.
Example Claude API costs for three workloads
| Workload per month | Haiku 5.5 | Sonnet 5.5 | Opus 5.5 |
|---|---|---|---|
| Support bot: 10,000 replies, 3,000 tokens in and 400 out each | $5 | $100 | $200 |
| Ticket summaries: 100,000 tickets, 1,500 in and 150 out each, standard | $22.50 | $450 | $900 |
| The same summaries through the Batch API | $11.25 | $225 | $450 |
Per request, that support reply costs $0.0005 on Haiku 5.5, one cent on Sonnet 5.5 and two cents on Opus 5.5. These examples count thinking in their output tokens and use the same counts on every model, but Sonnet 5.5 defaults to high effort, against medium on Haiku 5.5 and Opus 5.5, so Sonnet's real bill may run higher. For chat-sized requests, the Claude API is cheap. Bills climb when requests get long: agents resending big contexts at every step, whole documents in the prompt, or the top models writing long, heavily reasoned answers.
Cut the bill with prompt caching, batch and effort
Caching pays when the same content goes into many requests. Take a 40-step coding agent on Opus 5.5 whose context starts at 20,000 tokens, grows by 3,000 a step, and writes 1,500 tokens a step, thinking included:
- Without caching: 3.14 million input tokens ($12.56) plus 60,000 output tokens ($1.20): $13.76 a session.
- With caching: 3.0 million of those input tokens become cache reads ($0.60), 137,000 are cache writes ($0.69), and output stays $1.20: about $2.49, or 82% less.
The same goes for a document you keep asking about. Say you ask 20 questions about a 50,000-token contract on Sonnet 5.5 within an hour:
- Without caching: 20 ร 50,000 tokens at $2 per million = $2.00 for the contract's input.
- With a 1-hour cache: one write at $4 per million ($0.20) plus 19 reads at $0.10 per million ($0.095) = about $0.30, 85% less.
Cache reads cost 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 5.2. A 5-minute cache write pays off after one read; a 1-hour write pays off after two.
For work that can wait, like the ticket summaries above, use the Batch API. Most batches finish within an hour, requests still unfinished after 24 hours expire without charge, and the batch discount stacks with caching.
Before switching to a smaller model, lower the effort setting. Effort steers how much the model thinks, and Anthropic says tuning it is often a better lever than switching models. And set a monthly spend limit in the Console so a runaway loop can't surprise you.
Claude Code API pricing: API key or subscription?
Signed in with a Claude Console account, Claude Code bills every token at the API rates above. Anthropic's own figure across enterprise deployments is about $13 per developer per active day, or $150 to $250 per developer per month, with 90% of users under $30 a day.
Pro includes Claude Code for $20 a month, and Max and Ultra raise its usage limits. At Anthropic's average, API billing passes Pro's price in under two active days and Ultra's $200 after about 15.
- If you code with Claude most days: take a subscription. Start on Pro and move to Max if you keep hitting its five-hour or weekly limits.
- If you only use it on the odd day: use an API key. One average day costs less than a month of Pro.
- If you run it in CI and scripts, or for a team that needs spending caps: use an API key in a Console workspace with its own spend limit. A team on a Team or Enterprise plan can instead cap spend with seat allowances and spend limits per organization, group or member.
Two traps: an ANTHROPIC_API_KEY variable in your environment makes Claude Code bill that key instead of your plan, and Max, Ultra or Team API credits can't pay for interactive Claude Code; only credits you buy in the Console can.
On an API key, /usage shows what a session has cost, /clear between unrelated tasks keeps context small, and Sonnet as the default keeps rates low; Anthropic recommends saving Opus for complex architecture and multi-step reasoning. Claude Code pricing covers the plan side in full.
Is the Claude API free?
Only for a first test. Anthropic says new users receive free credits to try the API but doesn't publish the amount. After that you pay before you use: buy prepaid credits on the Billing page of the Claude Console, and each request draws them down. Every Claude model has a per-token price; none is free on the API. Max, Ultra and Team subscribers can spend their monthly API credits instead.
- Bought credits are available immediately, expire a year after purchase and can't be refunded. Auto-reload can top up your balance when it drops below a level you set.
- Spend caps come with your usage tier: $500 a month on Start, $1,000 on Build and $200,000 on Scale. New organizations may begin on a lower Evaluation tier. You can set a lower limit yourself, and requests stop when you reach it.
Haiku vs Sonnet vs Opus: which Claude model is cheapest for your task?
Haiku 5.5 costs one-twentieth of Sonnet 5.5 per token and one-fortieth of Opus 5.5. The cheapest model is the smallest one that still passes your own tests, so the pick depends on what you're building:
- Classifying, extracting or routing at volume: Haiku 5.5, which Anthropic built for high-volume, latency-sensitive work.
- A chatbot, a writing feature, everyday coding or another typical production app: Sonnet 5.5, Anthropic's suggestion for everyday coding, agent and enterprise workloads.
- Agents that run for hours, or complex coding: Opus 5.5, Anthropic's recommended starting point when you're unsure.
- Hard reasoning or long agent runs where Opus 5.5 falls short in your tests: Fable 5.2. Our Fable 5.1 review covers when the Fable line earns its price.
Weighing Haiku 4.5 against Sonnet 5? Sonnet 5 costs twice as much per token, but Haiku 5.5 undercuts both at a tenth of Haiku 4.5's rate for prompts under 100,000 tokens, and Anthropic advises defaulting to its latest models. You can also mix models: route simpler tasks to cheaper models and save Sonnet or Opus for the hard ones.
Claude API pricing vs OpenAI and Gemini
At matching price points the three providers are close. Sonnet 5.5 and OpenAI's GPT-6.2 Sol both list $2 in and $10 out, and Haiku 5.5 and GPT-6 Luna both list $0.10 and $0.50. Opus 5.5 at $4/$20 costs less than GPT-6.2 Astra at $8/$40. Long prompts change the comparison: GPT-6.2 Sol, GPT-6 Luna and GPT-6.2 Astra cost more above 272,000 tokens and Haiku 5.5 above 100,000, while Sonnet 5.5 and Opus 5.5 keep one rate to 1 million. Google's Gemini 4 Carbon and Argon list $2/$10 until November 30, 2026, then $4/$20.
Two things matter more than the rate card. Google offers a free tier on several Gemini models; Claude's API has nothing beyond the starting credits. And models use different numbers of tokens for the same job, so compare the billed cost of the same sample requests, not per-token prices. Opus 5.5 vs GPT-6 Astra and Google AI Studio pricing go deeper on each, and ChatGPT pricing covers the chat subscriptions.
Frequently Asked Questions
How much does the Claude API cost?
It depends on the model. Per million tokens, Claude Haiku 5.5 costs $0.10 for input and $0.50 for output, Sonnet 5.5 costs $2 and $10, Opus 5.5 costs $4 and $20, and Fable 5.2 costs $8 and $40. The Batch API halves those rates, and cached input costs a small fraction of the normal input price.
Is API access included in Claude Pro?
No. Pro includes Claude and Claude Code but no API credits. Max ($100 a month in credits), Ultra ($200) and Team (up to $500, pooled) include monthly API credits, and those credits can't be used for interactive Claude Code.
How much does Claude Haiku 4.5 cost?
Claude Haiku 4.5 costs $1 per million input tokens and $5 per million output tokens, or $0.50 and $2.50 through the Batch API. The newer Claude Haiku 5.5 costs $0.10 and $0.50 for prompts up to 100,000 tokens.
How much cheaper is Haiku than Sonnet and Opus?
Claude Haiku 5.5 is 20 times cheaper per token than Sonnet 5.5 and 40 times cheaper than Opus 5.5. The older Haiku 4.5 costs half as much as Sonnet 5.5 and a quarter as much as Opus 5.5.
Why is Claude so expensive?
On the API, cost climbs with output and context rather than with each question. Output costs five times as much as input, thinking is billed as output, chats and agents resend their history with every request, and Fable 5.2 costs 80 times more per token than Haiku 5.5. Prompt caching, the Batch API and a smaller model all cut the bill.
Which AI API is the cheapest?
At the budget end, for prompts under 100,000 tokens, Claude Haiku 5.5 and OpenAI's GPT-6 Luna both cost $0.10 per million input tokens and $0.50 per million output tokens, and OpenAI's older GPT-5 nano lists $0.05 and $0.40. Google's Gemini API also offers a free tier on several models.
Claude Haiku 4.5 vs GPT-5 mini: which costs less?
GPT-5 mini, per token: OpenAI lists it at $0.25 per million input tokens and $2 per million output tokens, against $1 and $5 for Claude Haiku 4.5. Claude Haiku 5.5 is cheaper per token than both, at $0.10 and $0.50 for prompts under 100,000 tokens.
Advertiser disclosure: some links on this website are affiliate links, meaning No Code MBA may make a commission if you click through and purchase.