Claude API Pricing 2026: Costs for Small Coaching Businesses

Claude API pricing is not the expensive part of using AI in a coaching business. Not knowing what each task costs is.

The API bills per token, separately for what you send and what Claude writes back. Most coaching tasks cost cents. The risk is not the per-task price. It is running workflows nobody measured, on a model bigger than the job needed, with no spending cap.

This guide gives you the current per-model prices as listed by Anthropic on 1 October 2026, the discounts that matter, worked cost examples with the arithmetic shown, an honest comparison with a Claude Pro or Team subscription, and the exact steps to cap your spend. The Apex Life Fitness editorial team is not affiliated with Anthropic. Prices change, so check the official pricing page before you budget.

Claude API pricing in 2026: the current models

Anthropic's models overview lists four current models. Prices are in US dollars per million tokens (MTok), as shown on the docs pricing page and the API tab of claude.com/pricing.

Model	Input	Output	Cache hit	Batch input / output
Claude Fable 5.1	$10	$50	$0.25	$5 / $25
Claude Opus 5.5	$4	$20	$0.20	$2 / $10
Claude Sonnet 5.5	$2	$10	$0.20	$1 / $5
Claude Haiku 4.5	$1	$5	$0.10	$0.50 / $2.50

Anthropic describes Fable 5.1 as the model for demanding reasoning and long-horizon agentic work, Opus 5.5 as the starting point for most workloads, Sonnet 5.5 as the best combination of speed and intelligence, and Haiku 4.5 as the fastest.

Older models are still billable. Claude Sonnet 5, for example, stays at $2 input and $10 output. The pricing page notes that this launch price is now standard and the previously scheduled increase will not happen. Several third-party pricing pages still lead with older models, which is one more reason to read the source.

What the columns mean

Input is everything you send: instructions, client notes, documents, the conversation so far.
Output is what Claude writes. It costs five times the input rate on every current model.
Cache hit is the price of re-reading a prompt prefix you already stored in the cache. More on this below.
Batch is the price when you submit work asynchronously through the Batch API: half of the standard rate.

Two modifiers can raise the price. Fast mode on Opus 5.5 is listed at $8 input and $40 output, double the standard rate. Requesting US-only inference applies a 1.1x multiplier to all token prices on Claude 4.6 and later models. A small coaching business rarely needs either.

How tokens translate into real work

A token is a fragment of text, not a word. Anthropic's pricing FAQ gives a general rule of about 0.75 words per token in English. The newer tokenizer used by Claude 4.7 and later models produces approximately 30% more tokens for the same text, according to the same page.

The models overview puts it concretely: on the current tokenizer, 1M tokens is roughly 555,000 words. That works out to about 1.8 tokens per word. The examples below use that figure, because Sonnet 5.5 and Opus 5.5 run on the newer tokenizer. Haiku 4.5 is not among the models on the newer tokenizer, so the same text produces fewer tokens there.

Two facts follow. Long inputs are cheap; long outputs are where the bill grows. And the model choice moves the price more than anything else you control.

Worked cost examples for a coaching business

These are estimates, not quotes. Each one states its assumptions so you can swap in your own numbers. Real token counts depend on your text; the Console reports the exact usage of every request.

Example 1: weekly client check-in replies on Sonnet 5.5

Assumptions: 1,500 words of coaching notes and instructions plus a 300-word client message as input, a 250-word reply as output, 30 clients, 4 check-ins each per month.

Input: 1,800 words × 1.8 = 3,240 tokens × $2 / 1,000,000 = $0.00648
Output: 250 words × 1.8 = 450 tokens × $10 / 1,000,000 = $0.0045
Per reply: about $0.011. Per month (120 replies): about $1.32

The same reply on Fable 5.1 costs about $0.055, five times as much. For a drafted check-in that a coach reviews before sending, that difference buys nothing.

Example 2: call transcript to session summary

Assumptions: a 6,000-word transcript plus 500 words of instructions as input, a 600-word summary as output, 40 calls per month, Sonnet 5.5.

Input: 6,500 words × 1.8 = 11,700 tokens × $2 / 1,000,000 = $0.0234
Output: 600 words × 1.8 = 1,080 tokens × $10 / 1,000,000 = $0.0108
Per summary: about $0.034. Per month: about $1.37

Run the same 40 summaries through the Batch API at $1 and $5 and the month drops to about $0.68. The trade is speed: Anthropic's batch processing docs say most batches finish within an hour, but results can take up to 24 hours.

Example 3: long-form content drafts on Opus 5.5

Assumptions: a 2,500-word brief and voice guide as input, an 1,800-word article draft as output, 8 drafts per month.

Input: 2,500 words × 1.8 = 4,500 tokens × $4 / 1,000,000 = $0.018
Output: 1,800 words × 1.8 = 3,240 tokens × $20 / 1,000,000 = $0.0648
Per draft: about $0.083. Per month: about $0.66

Output is 78% of that bill. Asking for a tighter draft saves more than trimming the brief.

The monthly total

All three workflows together come to about $3.35 a month at standard rates, or about $2.66 with the summaries batched. That is the token bill only. It does not include the time or the tool needed to connect the API to your inbox, CRM or forms, which is the real cost for a non-developer.

Prompt caching: the discount for repeated context

Most coaching prompts repeat the same opening: your method, your voice, your rules. Prompt caching stores that prefix so later requests re-read it at a fraction of the input price.

The multipliers, per the pricing page: a 5-minute cache write costs 1.25x the base input price, a 1-hour cache write costs 2x, and a cache hit costs 0.1x on most models. Opus 5.5 hits are 0.05x and Fable 5.1 hits are 0.025x. Caching also has a minimum prompt length; shorter prompts are processed without caching and no error is returned.

Example 4: a 5,000-word coaching playbook in every check-in

Assumptions: Sonnet 5.5, a 5,000-word playbook (about 9,000 tokens) added to each of 120 monthly check-ins, processed in 4 weekly sessions of 30 within one hour, using the 1-hour cache.

Without caching: 9,000 tokens × 120 × $2 / 1,000,000 = $2.16
With caching, per session: one write at $4 / MTok (9,000 × $4 / 1,000,000 = $0.036) plus 29 hits at $0.20 / MTok (29 × 9,000 × $0.20 / 1,000,000 = $0.0522) = $0.0882
Four sessions: about $0.35

The saving only appears if the requests land inside the cache window. Scattered requests across a week will mostly pay full price. Caching rewards a schedule, not a habit.

API or a Claude Pro or Team subscription?

This is the question a non-developer coach actually needs answered. The two are separate products with separate bills. Anthropic's help centre states that a paid Claude subscription does not include access to the Claude API or Console.

The subscription prices listed on claude.com/pricing today:

Pro: $20 per month billed monthly, or $17 per month with the annual plan ($200 billed up front).
Max: from $100 per month, with 5x or 20x more usage than Pro.
Team: for 2 to 150 people. Standard seats are $20 per seat per month billed annually or $25 monthly; premium seats are $100 annually or $125 monthly.

Subscriptions carry usage limits rather than a token bill. The API carries no limit on what you can build, and a bill for every token.

Choose the subscription when

You work with Claude in a chat window: drafting, thinking, reviewing documents.
You want Projects, file uploads and the apps without writing code. Our guide to setting up Claude Projects covers the coaching setups.
You want a predictable monthly cost.

Choose the API when

A task should run without you: a form submission triggers a draft, a transcript triggers a summary.
You need Claude inside another tool, through an automation platform or your own code.
The volume is high and repetitive, where batching and caching cut the price.

Many small businesses need both: a subscription for the daily thinking work, a capped API account for one automated workflow. On the examples above, the token bill for that workflow is a fraction of one Pro seat. If you are still choosing an assistant at all, start with the Claude vs ChatGPT comparison.

How to cap your Claude API spend

The rate limits documentation describes two layers of control.

First, every usage tier carries a monthly spend cap: $500 on the Start tier, $1,000 on Build and $200,000 on Scale. When an organization reaches its tier cap, API usage pauses until 00:00 UTC on the first day of the next month unless a higher limit is requested.

Second, you can set your own limit below that cap. In the Claude Console, go to Settings, then Billing, and in the Spend limits section choose Adjust limit or Set limit. When usage reaches your limit, requests return an error stating when access resumes. Workspaces can also carry their own spend and rate limits, though not the default Workspace.

New accounts receive a small amount of free credits to test the API, according to the pricing FAQ. Use them to measure your real token counts before you set the budget.

The operator protocol: price a workflow before you run it

Name one workflow. One input, one output, one owner. "Draft the weekly check-in reply" qualifies. "Use AI for the business" does not.
Pick the smallest model that passes. Start on Sonnet 5.5 or Haiku 4.5. Move up only when a reviewed sample of outputs fails your standard.
Measure ten real runs. Read the input and output token counts from the Console usage view. Replace the 1.8 tokens-per-word estimate with your own average.
Price the month. Tokens per run × price per million ÷ 1,000,000 × runs per month. Write the number down.
Set the spend limit at two to three times that number. High enough to absorb a busy month, low enough that a broken loop cannot become a large invoice.
Apply the discounts that fit. Batch anything that can wait a day. Cache any prefix reused within the window.
Review monthly. Compare actual spend with the estimate. A gap means the workflow changed or something is looping.
Keep a human check. Claude drafts; the coach approves anything a client reads. The seven daily small-business workflows show where that line sits.

Common mistakes with Claude API costs

Defaulting to the largest model. Fable 5.1 costs five times Sonnet 5.5 per token. Use it where the reasoning requires it, not for routine drafts.
Budgeting on word count. On the current tokenizer, a word is closer to 1.8 tokens than 1. Underestimate tokens and you underestimate the bill.
Ignoring output length. Output costs five times input. Specify a length in every prompt.
Resending the whole history. Every turn of a long conversation re-bills everything before it as input. Summarise and restart.
No spend limit. An automation that retries on error can run all night. A limit turns that into a pause instead of an invoice.
Trusting third-party price tables. Many still list older models. Anthropic's own page is the only source that counts.
Forgetting tool costs. Web search on the API is billed at $10 per 1,000 searches on top of tokens, and tool definitions add input tokens to every request.

Frequently asked questions

How much does the Claude API cost per month for a small coaching business?

There is no monthly fee; you pay per token. Our estimate for three typical workflows on Sonnet 5.5 and Opus 5.5 (check-in replies, call summaries and eight content drafts) is about $3.35 a month at standard rates. Your number depends on volume, model and output length, so measure ten real runs first.

Is Claude API access included in Claude Pro?

No. Anthropic's help centre states that a paid Claude subscription, including Pro, Max, Team and Enterprise, does not include access to the Claude API or Console. The API is set up and billed separately in the Claude Console, per token, alongside any subscription you hold.

Which Claude model is cheapest on the API?

Among the current models, Claude Haiku 4.5 is the cheapest at $1 per million input tokens and $5 per million output tokens, or half that through the Batch API. Sonnet 5.5 costs twice as much per token and is the practical default for most coaching drafts.

How do I stop the Claude API from overspending?

Set a spend limit in the Claude Console under Settings, then Billing, in the Spend limits section. Your limit must sit below your usage tier's monthly cap. When usage reaches it, requests stop with an error until access resumes, so a broken automation cannot keep billing you.

What is the Claude Batch API discount?

The Batch API processes requests asynchronously at a 50% discount on both input and output tokens. Anthropic says most batches finish within an hour, but results can take up to 24 hours. It suits work that can wait, such as overnight summaries or bulk content tagging.

The real cost is attention

The token bill for a well-scoped coaching workflow is small. The expensive version is the one nobody measured, running on the wrong model, with no limit. Price it, cap it, review it. That is discipline applied to a tool.

The same principle governs training and thinking: systems compound, intentions do not. If you want the daily structure that holds all three together, start the free 5-Day Stoic Operator Challenge.

Claude API Pricing 2026: What It Costs a Small Coaching Business