AI Token Pricing Explained: How to Estimate and Control What AI Costs You

The short answer

AI token pricing charges you for the text a model reads (input tokens) and writes (output tokens), priced per million tokens, with output costing about five times more than input on most current models. To estimate a monthly bill, measure the tokens in one typical task, multiply by your monthly volume, and apply the published rate. The same job can cost a hundred times more on a flagship model than on a small one, so model choice is the biggest cost lever you have.

Seat-based plans like ChatGPT Business or Claude Team hide tokens from you: you pay per person per month. The moment you automate something, such as answering emails, summarising calls or running an AI agent, you usually move to an API, and the bill becomes metered. This guide explains AI token pricing in plain terms, with real October 2026 rates from the official pricing pages and a worked example you can copy.

What a token is, and why the count varies

A token is a chunk of text, often part of a word. Models read and write in tokens, and vendors bill in them. The catch: different models split the same text differently. Anthropic's pricing documentation notes that Claude 4.7 and later models use a newer tokenizer that "produces approximately 30 percent more tokens for the same text" than earlier models. So a lower price per token doesn't always mean a lower price per task.

The practical fix is to measure. Most providers offer a token counter in their console or API. Run five real examples of your task, note the input and output tokens, and average them. That number is the foundation of every estimate below.

AI token pricing in October 2026: the published rates

All prices are in US dollars per million tokens, standard tier, from each vendor's official pricing page as of October 2026.

Model	Input	Output	Notes
OpenAI GPT-6 Luna	$0.10	$0.50	Efficient tier
OpenAI GPT-6.1 Sol	$2.00	$10.00	Higher rates above 272K input tokens
OpenAI GPT-6 Astra	$10.00	$50.00	Most capable OpenAI model
Claude Haiku 5.5	$0.10	$0.50	Rises to $0.50 / $2.50 for prompts over 100,000 tokens
Claude Sonnet 5.5	$2.00	$10.00	Full 1M context at standard price
Claude Opus 5.5	$4.00	$20.00	Full 1M context at standard price
Gemini 3.5 Flash-Lite	$0.30	$2.50	Free tier available
Gemini 4 Argon	$2.00	$10.00	Introductory; $4 / $20 from December 1, 2026

Sources: OpenAI API pricing, Anthropic pricing, Gemini Developer API pricing. One date to diary: Google's own Gemini 4 Argon announcement says the introductory rate ends November 30, 2026, which doubles the price of anything you build on it today.

Context windows: why long conversations get expensive

The context window is how much text a model can consider at once. OpenAI lists a 1.05M-token window for its GPT-6 flagships, and Google lists a 1M-token limit for Gemini 4 Argon. That's hundreds of pages.

Here's the part that surprises people: you pay for the whole context every time. In a chat or an AI agent, each new turn usually resends the earlier conversation, the instructions and any attached documents. A ten-step agent working over a 50-page contract can process that contract ten times. Long context is useful; it's also where bills quietly grow.

Watch the thresholds too. OpenAI charges more once a prompt exceeds 272K input tokens; for GPT-6 Astra, input doubles from $10 to $20 and output rises from $50 to $75. Claude Haiku 5.5 switches to its higher rate above 100,000 tokens. Anthropic says Sonnet 5.5 and Opus 5.5 bill their full 1M window at the standard price.

A worked example: 3,000 customer emails a month

Suppose you want AI to draft replies to 3,000 support emails a month. You measured your task and found each one uses about 2,000 input tokens (instructions, your FAQ snippet and the email) and 400 output tokens (the draft). Those are assumptions for illustration; measure your own.

• Monthly input: 3,000 × 2,000 = 6 million tokens
• Monthly output: 3,000 × 400 = 1.2 million tokens

At $0.10 input and $0.50 output (GPT-6 Luna or Claude Haiku 5.5), that's $0.60 + $0.60 = about $1.20 a month. At $2 and $10 (Claude Sonnet 5.5, GPT-6.1 Sol, Gemini 4 Argon's introductory rate), it's $12 + $12 = $24. At $10 and $50 (GPT-6 Astra), it's $60 + $60 = $120.

None of these numbers would sink a business. That's the real lesson: for most small-business workloads, token costs are small next to the cost of building, testing and maintaining the automation. Our breakdown of what AI automation really costs covers those other lines, and our chatbot cost guide does the same for website assistants.

Plan tiers versus API billing

Two billing models exist side by side, and they suit different jobs.

• Seat plans are a flat monthly fee per person for a chat app. Anthropic lists Claude Team Standard at $25 per seat billed monthly. Usage limits apply, but you don't count tokens. Good for staff using AI interactively.
• API billing is metered per token, with no seat fee. Good for automations, agents and anything running without a person in the loop.
• Hybrid enterprise plans combine both. Anthropic's Enterprise plan is listed at $20 per seat annually plus usage at API rates.

Choosing between vendors? Our ChatGPT vs Claude vs Gemini comparison covers plans and data terms side by side.

Six ways to keep AI costs under control

1. Route by difficulty. Send simple tasks (classifying, extracting, short replies) to a small model and escalate only the hard ones. Our guide to small language models explains when a small model is enough.
2. Use prompt caching. If every request starts with the same long instructions, caching cuts that cost sharply. Anthropic prices cache hits on Opus 5.5 and Sonnet 5.5 at 0.05 times the base input price; Google lists cached input on Gemini 4 Argon at 95% off.
3. Batch what can wait. Anthropic's Batch API gives "a 50% discount on both input and output tokens", and OpenAI and Google list batch rates at about half of standard.
4. Trim the context. Send the relevant pages, not the whole file. Retrieval, which we explain in our RAG glossary entry, exists partly for this reason.
5. Cap output length. Output tokens cost about five times more than input on these models. Ask for the length you need.
6. Set a spending cap before launch. Check your provider console for usage limits and alerts, and test with a cap in place so a looping agent can't run up a surprise bill.

One more line item to know about: data residency. OpenAI adds 10% for regional processing, and Anthropic applies a 1.1x multiplier for US-only inference. If a client contract requires a specific region, budget for it.

Key takeaways

• You pay separately for input and output tokens; output costs roughly five times more on current models.
• Measure tokens on real examples, then multiply by monthly volume. Guesses are usually wrong.
• Model choice swings cost by about 100 times for the same task; routing easy work to small models is the biggest saving.
• Caching, batching and shorter context are the next three levers.
• Watch dated changes, like Gemini 4 Argon's price doubling on December 1, 2026.

Frequently asked questions

How many tokens is a word?

It depends on the model's tokenizer and the language, and it differs between vendors. Anthropic says its newer tokenizer produces about 30% more tokens for the same text than its older one. Count tokens on your own examples rather than relying on a rule of thumb.

Is ChatGPT Plus or the OpenAI API cheaper for a business?

For people chatting with AI during the day, a seat plan is usually simpler and more predictable. For automations that run without a person, the API is the right tool, and at small-business volumes it's often a few dollars to a few hundred dollars a month.

What is a context window in AI?

It's the amount of text a model can consider in one request, including your instructions, documents and the conversation so far. As of October 2026, OpenAI's GPT-6 flagships list 1.05M tokens and Gemini 4 Argon lists 1M. You pay for everything you put in it.

Sources

1. Pricing — OpenAI API documentation, accessed October 17, 2026
2. Pricing — Anthropic (Claude Platform docs), accessed October 17, 2026
3. Gemini Developer API pricing — Google AI for Developers, accessed October 17, 2026
4. Gemini 4 Argon: Built to solve your hardest problems — Google, October 13, 2026
5. Models — OpenAI API documentation, accessed October 17, 2026
6. Pricing — Anthropic (Claude plans), accessed October 17, 2026

How Flowmatix can help

We put this into practice for businesses in Toronto and around the world through our AI Agents & AI Solutions, business process automation services. Want to know what it would take for your business? Get a personalised quote in about 3 minutes.