How much do AI tokens cost? 1M and 10M token prices in 2026

You see “$0.50 per 1M tokens” and have no idea what that means for your app. The short version: multiply each price by how many millions of tokens you use, once for input and once for output, then add the two. For the eight models in the table below (official pricing pages, checked 26 October 2026), 1M input tokens cost from $0.10 to $4.00, and a month with 10M input and 2M output tokens comes to between $2.00 and $80.00.

The numbers come from the pricing pages of OpenAI, Anthropic and Google. Prices as read on 26 October 2026; the pages carry no publication date. Live CheapTokens numbers come further down.

The formula, on a real month#

Cost = input millions × input price + output millions × output price. That is the whole trick. If a token is still a fuzzy idea, what is a token is the plain-English version, and input vs output tokens explains why the two sides have different prices.

Say you run a small support bot. It handles 10,000 chats a month; each chat sends about 1,000 tokens in (the customer's message plus your instructions) and gets about 200 tokens back. That is 10,000 × 1,000 = 10,000,000 input tokens and 10,000 × 200 = 2,000,000 output tokens: the 10M + 2M month used in the rest of this guide.

Price it on OpenAI's GPT-6 Luna, 0.10 input and 0.50 output per 1M tokens: 10 × 0.10 + 2 × 0.50 = 2.00 dollars. Look at the two halves. Input costs $1.00, output costs $1.00. The answer is five times shorter than the question, and it still costs the same. That is the output price at work, and it is why you read both columns, not just the first.

Official prices, 26 October 2026#

USD per 1M tokens, standard tier, base input price. The last column is the 10M + 2M month: millions times price, added up.

Four catches before you trust any row:

• Claude Haiku 5.5 has a step. It is 0.10 / 0.50 for prompts up to 100,000 tokens and 0.50 / 2.50 above that. A long-document month at the higher tier: 10 × 0.50 + 2 × 2.50 = 10.00.
• Some prices have an end date. Gemini 3.6 Flash is on 0.75 / 3.75 until 31 December 2026, then 1.50 / 7.50, and the same month becomes 10 × 1.50 + 2 × 7.50 = 30.00. Gemini 4 Argon moves on 1 December 2026, per Google's page: from 2.00 / 10.00 (10 × 2.00 + 2 × 10.00 = 40.00) to 4.00 / 20.00 (10 × 4.00 + 2 × 20.00 = 80.00).
• Gemini counts thinking tokens as output. A reasoning model can bill more output than the answer you see on screen; reasoning effort and thinking tokens shows how that adds up.
• Cached input is priced differently by each vendor. The table uses base input only; what is prompt caching covers the discount side.

The same month on CheapTokens#

CheapTokens bills the same way: per 1M tokens, input and output separately, from one balance for all models, through an API compatible with OpenAI and Anthropic. Today's price for MiniMax-M2.7, read live from the price list: $0.05 in / $0.15 out per 1M tokens.

The support-bot month, 10M in + 2M out, next to a few official rows from the table above:

And the full live table for the same month, with OpenRouter next to it:

MiniMax-M2.7 is cheaper than OpenRouter in total, with input and output counted together. Current prices for every model are on the pricing page.

Which model for the support bot#

The support bot from above is short questions and short answers, which suits MiniMax-M2.7: 180K context, up to 8,192 output tokens per answer, cheaper than OpenRouter in total, and one price at any hour, with no peak times.

Check your own bill with one request#

Send one real request, then read the usage block in the reply. It carries prompt_tokens (input) and completion_tokens (output). Those are the two numbers from the formula for a single request; multiply by your monthly request count and you have your own version of 10M + 2M. How many tokens an API request uses walks through three measured tasks.

curl https://api.cheaptokens.dev/v1/chat/completions \
  -H "Authorization: Bearer YOUR_API_KEY" \
  -H "Content-Type: application/json" \
  -d '{"model": "MiniMaxAI/MiniMax-M2.7", "messages": [{"role": "user", "content": "Summarize in one line: the order shipped late."}]}'

Model ids differ between providers, so take the one for your model from the CheapTokens docs. Run a day of real traffic and look at usage: your own ratio of input to output is the number that decides the bill.