See what an AI feature will cost before you build it. Enter your traffic and tokens to compare Claude, GPT and Gemini side by side, with caching and batch discounts.
Share of each prompt that's cached
If every request starts with the same instructions or documents, providers charge much less to reuse them. 50 to 90% is common for chatbots with a long system prompt.
About 0 tokens
A rough guide for English: about 4 characters or ¾ of a word per token. Code and other languages often use more. Nothing you paste leaves your browser.
Claude Sonnet 5.5 would cost about
$1,200/month
Cheapest here: Claude Haiku 5.5 at $60.00/month. The most expensive costs about 80× as much. Cheaper only matters if it passes your tests.
| Model | Tier | Input | Cached input | Output |
|---|---|---|---|---|
| Claude Fable 5.2Anthropic | Frontier | $8 | $0.20 | $40 |
| GPT-6.2 AstraOpenAI | Frontier | $8 | $0.40 | $40 |
| Gemini 4 CarbonGoogle | Frontier | $4 | $0.20 | $20 |
| Gemini 4 ArgonGoogle | Frontier | $4 | $0.20 | $20 |
| Claude Opus 5.5Anthropic | Everyday | $4 | $0.20 | $20 |
| Gemini 4 NeonGoogle | Everyday | $3 | $0.15 | $15 |
| Claude Sonnet 5.5Anthropic | Everyday | $2 | $0.10 | $10 |
| GPT-6.2 SolOpenAI | Everyday | $2 | $0.10 | $10 |
| Gemini 4 HydrogenGoogle | Everyday | $1.50 | $0.075 | $7.50 |
| GPT-6.1 TerraOpenAI | Everyday | $1 | $0.05 | $5 |
| Gemini 3.5 Flash-LiteGoogle | Small | $0.30 | $0.03 | $2.50 |
| Gemini 3.1 Flash-LiteGoogle | Small | $0.25 | $0.025 | $1.50 |
| Claude Haiku 5.5Anthropic | Small | $0.10 | $0.01 | $0.50 |
| GPT-6 LunaOpenAI | Small | $0.10 | $0.01 | $0.50 |
Cost per request is the uncached input tokens times the input price, plus cached tokens times the cached price, plus output tokens times the output price, all divided by a million. Batch halves the total. A month is 30 days.
Prices are list prices for short prompts. Very long prompts can cost more on some models, writing to the cache has its own fee, and Gemini 4 Carbon and Argon are shown at their prices from December 1, 2026. Always check the provider's pricing page before you set a budget.
Everything runs in your browser. Nothing you type or paste is sent anywhere.
Multiply input tokens by the input price and output tokens by the output price, add them, and divide by one million, since prices are quoted per million tokens. Multiply by your number of requests to get a daily or monthly total.
A token is a chunk of text the model reads or writes. In English it's roughly four characters or three quarters of a word, so 1,000 tokens is about 750 words. Code, numbers and many other languages use more tokens per word.
The model has to generate output one token at a time, which takes much more computing than reading the prompt. On most current models output costs about five times as much as input, so long replies drive the bill.
Cached input usually costs between 2.5% and 10% of the normal input price. If most of every prompt is the same instructions or documents, caching can cut input costs by more than half. Writing to the cache has a small extra fee that this calculator leaves out.
Anthropic, OpenAI and Google all offer a batch API at half price. You send many requests at once and get results back within about a day instead of instantly. It suits work nobody is waiting on, like nightly reports or labelling old data.
In November 2026 the cheapest models here are Claude Haiku 5.5 and GPT-6 Luna at $0.10 per million input tokens and $0.50 per million output tokens. The cheapest model is only a good deal if it passes your own tests.