Tokens, Context Windows and API Pricing Explained: Claude vs ChatGPT vs Gemini (2026)

Claude vs ChatGPT vs Gemini · Part 6 of 10

In this post we are going to cover tokens, context windows and API pricing for Claude, ChatGPT and Gemini. If you build anything on these models, this is how your bill is calculated.

The idea is simple: you pay for the text that goes in and the text that comes out, counted in tokens. The details are where the money is. We go through seven questions, each with the short answer first and then the numbers.

Every price and limit in this post was checked on the official pricing and documentation pages on November 16, 2026. These numbers change often, so use the links at the end of the post to check them again before you pay for anything.

Part 6 of the Claude vs ChatGPT vs Gemini: The Complete 2026 Comparison series on TechBrothersIT.

1. What is a token, and why do prices use tokens?

The short answer

A token is a small piece of text, about three quarters of a word. The model reads tokens and writes tokens, and you pay for both. Output tokens cost about five times more than input tokens.

A model does not read words. It reads tokens. Take a short sentence. The model splits it into pieces. Short words are one token, and longer words are split into two or three. As a rule of thumb, 100 tokens is about 75 words. So 1 million tokens is about 750 thousand words.

The data pipeline failed at midnight.
100 tokens
about 75 words
1M tokens
about 750,000 words

Now the price. Text you send is input. Text the model writes is output. On the current Claude models and the GPT-6 models, output costs exactly 5 times the input price. So a long answer costs much more than a long question.

Input tokensyour prompt, files, history
The model
Output tokensthe answer · about 5x the price

One more thing. Each company counts tokens in its own way, so the same page of text can be a different number of tokens on each model. Anthropic says its newer models produce about 30 percent more tokens for the same text than its older ones.

Keep in mind

Token counts are not identical across models. Compare the bill, not only the price per token.

2. How big is the context window on the API?

The short answer

On the API, the three flagships are almost the same: 1M tokens on Claude, 1.05M on OpenAI, and 1M on Gemini. The maximum answer length is 128K tokens on Claude and OpenAI.

The context window is the total the model can hold: your prompt, the files, the conversation so far, and the answer. Claude's current models all have a 1 million token window. OpenAI's GPT-6 models have 1.05 million. Gemini 4 Argon has 1 million, and it accepts video and audio in that window too.

APIClaudeChatGPTGemini
FlagshipFable 5.2GPT-6 AstraGemini 4 Argon
Context window1M tokens1.05M tokens1M tokens
Max output128K tokens128K tokens—
Input typestext, images, PDFstext, images, filestext, images, video, audio, PDFs
Sources: Anthropic model docs, OpenAI pricing as reported by BenchLM, Google model card for Gemini 4 Argon

There is also a limit on the answer. Claude and OpenAI both allow up to 128 thousand tokens of output in one response. So the window is about what goes in, and the output limit is about what comes back.

Keep in mind

Window = what fits in. Max output = how long one answer can be.

3. Why is the context window smaller in the app than on the API?

The short answer

Because in the app you pay a flat price, so the company limits how much each message can use. Claude and Gemini list the full 1M in the app. ChatGPT lists 54K to 400K, depending on the plan.

On the API you pay for every token, so the company lets you use the whole window. In the app you pay a fixed price, so there is a cap. Claude lists up to 1 million tokens in the app, on every plan. Gemini lists 1 million on AI Pro. ChatGPT lists 54 thousand for instant answers on Plus, 256 thousand for reasoning, and 400 thousand on Pro. The same model has 1.05 million on the API.

Claude app
up to 1M
Gemini app · AI Pro
1M
ChatGPT app · Pro
400K
ChatGPT app · Plus
256K
GPT-6 on the API
1.05M
Sources: the three official pricing pages and the API documentation, checked November 16, 2026

So if a file is too long for ChatGPT Plus, the model itself is not the problem. The plan is. The same model reads the whole file through the API.

Keep in mind

Too long for ChatGPT Plus? The API version of the same model has a 1.05M window.

4. What do the top models cost per million tokens?

The short answer

From cheapest to most expensive: Gemini 4 Argon at $4 in and $20 out from December 1, Claude Fable 5.2 at $8 and $40, and GPT-6 Astra at $10 and $50.

Here are the three flagships. GPT-6 Astra: 10 dollars for a million input tokens, and 50 dollars for a million output tokens. Claude Fable 5.2: 8 and 40. That is 20 percent less than the previous Fable. Gemini 4 Argon: 4 and 20 from December 1. Until November 30 it is on an intro price of 2 and 10.

GPT-6 Astra
$10 in / $50 out
Claude Fable 5.2
$8 in / $40 out
Claude Opus 5.5
$4 in / $20 out
Gemini 4 Argon
$4 in / $20 out
Bars show the output price per 1M tokens. Standard rates from the official API pricing pages, November 2026. Argon is $2 / $10 until November 30.

Notice Claude Opus 5.5. It is Claude's second model, and it costs the same as Gemini 4 Argon: 4 and 20. Anthropic itself recommends starting with Opus for most work, and moving up to Fable only when you need it.

Keep in mind

Do not start with the most expensive model. Start one tier down and move up only if the results fall short.

5. How much cheaper are the smaller models?

The short answer

A lot. The mid models cost about $2 in and $10 out. The small models cost $0.10 in and $0.50 out, which is 100 times cheaper than GPT-6 Astra.

Here is the ladder. Mid models: Claude Sonnet 5.5 and GPT-6.1 Sol are both 2 and 10. GPT-6.1 Terra is 1 and 5. Gemini 3.8 Flash is 75 cents and 3 dollars 75 until the end of the year. Small models: Claude Haiku 5.5 and GPT-6 Luna are both 10 cents and 50 cents.

input / output per 1MClaudeChatGPTGemini
Mid modelSonnet 5.5 · $2 / $10GPT-6.1 Sol · $2 / $103.8 Flash · $0.75 / $3.75
Lower mid—GPT-6.1 Terra · $1 / $53.8 Flash doubles Jan 1
Small modelHaiku 5.5 · $0.10 / $0.50GPT-6 Luna · $0.10 / $0.50—
Standard rates from the official API pricing pages, November 2026

Let's make it real. Say you run 10 thousand support chats a month. Each chat sends 2 thousand tokens and gets back 400. That is 20 million input tokens and 4 million output tokens. On GPT-6 Astra, that costs 400 dollars. On Claude Fable 5.2, 320. On Gemini 4 Argon, 160. On a mid model, 80 dollars. And on a small model, 4 dollars.

GPT-6 Astra
$400
Claude Fable 5.2
$320
Gemini 4 Argon (Dec 1 rate)
$160
Sonnet 5.5 / GPT-6.1 Sol
$80
Haiku 5.5 / GPT-6 Luna
$4
10,000 chats x (2,000 input + 400 output tokens) = 20M input + 4M output tokens, at standard rates

Same job, from 400 dollars down to 4. For simple support questions, a small model is usually enough.

Keep in mind

Model choice changes the bill more than any discount. Use the smallest model that does the job.

6. Are there any hidden costs in these prices?

The short answer

Three. OpenAI charges more for long prompts, above 272K tokens. Two Gemini prices are introductory and rise on December 1 and January 1. And reasoning tokens count as output, even though you never see them.

Trap one: long prompts. On OpenAI's GPT-6 models, a request over 272 thousand input tokens is billed at twice the input price, and one and a half times the output price, for the whole request. On Claude, only Haiku 5.5 has a long prompt rate, above 100 thousand tokens. Gemini 4 Argon has one flat rate, but the older Gemini 3.1 Pro doubles above 200 thousand.

ClaudeChatGPTGemini
Long-prompt surchargeHaiku 5.5 only, over 100KGPT-6 models, over 272KArgon: none · 3.1 Pro over 200K
What changes$0.10 / $0.50 → $0.50 / $2.50input x2, output x1.53.1 Pro: $2 / $12 → $4 / $18
Flagship at long lengthFable 5.2: $8 / $40 (flat)Astra: $20 / $75Argon: $4 / $20 (flat)
Sources: the official API pricing pages, November 2026

Trap two: intro prices. Gemini 4 Argon doubles on December 1, from 2 and 10 to 4 and 20. Gemini 3.8 Flash doubles on January 1, from 75 cents and 3 dollars 75, to 1 dollar 50 and 7 dollars 50. Budget with the later price.

Gemini
Gemini 4 Argon
Until Nov 30: $2 / $10
From Dec 1: $4 / $20
Gemini
Gemini 3.8 Flash
Until Dec 31: $0.75 / $3.75
From Jan 1: $1.50 / $7.50

Trap three: thinking. When a model reasons before it answers, those hidden reasoning tokens are billed as output. A short answer can still cost a lot if the model thought for a long time.

Keep in mind

Reasoning tokens are billed as output. Lower the reasoning effort for simple tasks.

7. How can I pay less for the same work?

The short answer

Use two features that all three offer. Prompt caching makes repeated input up to 90 to 95 percent cheaper. Batch processing gives 50 percent off when you can wait for the answer.

First, caching. If you send the same long instructions or the same document again and again, the provider can cache it. On Claude Opus 5.5, cached input costs 20 cents instead of 4 dollars. On GPT-6.1 Sol, 10 cents instead of 2 dollars. On Gemini 4 Argon, 20 cents instead of 4 dollars, at the December rate.

per 1M tokensClaudeChatGPTGemini
ModelOpus 5.5GPT-6.1 SolGemini 4 Argon
Normal input$4.00$2.00$4.00 from Dec 1
Cached input$0.20$0.10$0.20
Batch, 50% off$2 / $10$1 / $5$2 / $10
Sources: the official API pricing pages. Claude also charges a one-time cache write fee.

Second, batch. If the job can wait, for example a nightly classification of 50 thousand records, send it as a batch. All three give half price. So the checklist is: pick the smallest model that works, cache what repeats, and batch what can wait. In the next video, we compare the coding tools.

Keep in mind

1. Smallest model that works 2. Cache what repeats 3. Batch what can wait

Key points to remember

  • A token is about three quarters of a word. You pay for input and for output, and output costs about 5 times more
  • On the API, all three flagships read about 1M tokens. In the app, ChatGPT gives much less
  • Flagships per 1M tokens: Gemini 4 Argon $4 / $20, Claude Fable 5.2 $8 / $40, GPT-6 Astra $10 / $50
  • Small models are 100 times cheaper: Haiku 5.5 and GPT-6 Luna are $0.10 / $0.50
  • Watch the 272K surcharge on OpenAI, and the Gemini prices that rise on December 1 and January 1

Sources

These are the official pages the numbers came from, read on November 16, 2026. Benchmark scores are named in the post together with who published them.

Next in this series

Part 7: Coding: Claude Code vs Codex vs Antigravity

This is an independent comparison. Prices are US list prices before tax, read on November 16, 2026. Plans, limits and models change often. Benchmark scores are reported as published, and each one names who ran it.

No comments:

Post a Comment

Note: Only a member of this blog may post a comment.