Free tool
LLM API Pricing Calculator
See what a month of requests costs on Claude, GPT, Gemini and DeepSeek models, from the tokens in one request, with prompt caching and batch discounts.
Runs in your browser: nothing you type is sent to us.
What 30,000 requests a month cost
The cheapest is GPT-5 nano at $6.04 a month, and the most expensive is GPT-6 Astra at $847.50 (140.4 times more).
That is 45 million input tokens and 12 million output tokens a month.
| Model | Inputper million | Cachedper million | Outputper million | Per request | Per month |
|---|---|---|---|---|---|
| GPT-5 nanoOpenAICheapest | $0.05 | $0.005 | $0.40 | $0.00020 | $6.04 |
| Gemini 2.5 Flash-LiteGoogle | $0.10 | $0.01 | $0.40 | $0.00024 | $7.281.2 times the cheapest |
| Claude Haiku 5.5Anthropic | $0.10 | $0.01 | $0.50 | $0.00028 | $8.481.4 times the cheapest |
| GPT-6 LunaOpenAI | $0.10 | $0.01 | $0.50 | $0.00028 | $8.481.4 times the cheapest |
| DeepSeek FlashDeepSeek | $0.30 | $0.006 | $1.20 | $0.00071 | $21.293.5 times the cheapest |
| Gemini 3.1 Flash-LiteGoogle | $0.25 | $0.025 | $1.50 | $0.00081 | $24.194 times the cheapest |
| Gemini 3.8 FlashGooglePrice changes on January 1, 2027 | $0.75 | $0.075 | $3.75 | $0.0021 | $63.5610.5 times the cheapest |
| DeepSeek V4 ProDeepSeek | $1.32 | $0.044 | $3.96 | $0.0026 | $78.2113 times the cheapest |
| GPT-6.1 TerraOpenAI | $1 | $0.05 | $5 | $0.0028 | $83.6313.9 times the cheapest |
| Claude Sonnet 5.5Anthropic | $2 | $0.10 | $10 | $0.0056 | $167.2527.7 times the cheapest |
| GPT-6.1 SolOpenAI | $2 | $0.10 | $10 | $0.0056 | $167.2527.7 times the cheapest |
| Gemini 4 ArgonGooglePrice changes on December 1, 2026 | $2 | $0.10 | $10 | $0.0056 | $167.2527.7 times the cheapest |
| Gemini 3.1 Pro (preview)Google | $2 | $0.20 | $12 | $0.0065 | $193.5032 times the cheapest |
| Claude Opus 5.5Anthropic | $4 | $0.20 | $20 | $0.011 | $334.5055.4 times the cheapest |
| Claude Fable 5.1Anthropic | $10 | $0.25 | $50 | $0.028 | $830.63137.6 times the cheapest |
| GPT-6 AstraOpenAI | $10 | $1 | $50 | $0.028 | $847.50140.4 times the cheapest |
- Thinking or reasoning tokens are billed as output, so a model that thinks first can use many more output tokens than the answer you see. Raise the output figure to match.
- Cache write fees, cache storage fees and the higher prices some providers charge for very long prompts are not counted. Each provider's notes are below.
- Prices are list prices in US dollars, before tax, volume deals and credits.
Where the prices come from
Prices checked on October 29, 2026, from each provider's own pricing page. They change often, so confirm on the page before you commit:
Anthropic
- Writing to the prompt cache costs 25% more than normal input for 5 minutes of caching (and 2 times for 1 hour). That one-off write fee is not in the numbers.
- The Batch API takes 50% off input and output.
- Claude Fable 5.1: Anthropic's most capable model.
- Claude Haiku 5.5: These prices are for prompts up to 100,000 tokens. Over that, input is $0.50 and output $2.50 per million.
OpenAI
- Caching is automatic and has no write fee.
- The batch tables on OpenAI's page are half the standard prices.
- Prompts over 272,000 tokens cost more per token (about 2 times for input and 1.5 times for output on most models). That is not in the numbers.
- GPT-6 Astra: OpenAI's top model.
- GPT-5 nano: An older, very small model.
Google
- Paid tier, standard prices. Cached input is also charged for storage by the hour, which is not in the numbers.
- The batch API is half the standard price.
- Some models charge more for prompts over 200,000 tokens.
- Gemini 4 Argon: Google has announced that the price doubles on 1 December 2026, to $4 input and $20 output per million. The calculator switches on that day.
- Gemini 3.8 Flash: Google has announced that the price doubles on 1 January 2027, to $1.50 input and $7.50 output per million. The calculator switches on that day.
- Gemini 3.1 Pro (preview): A preview model. These prices are for prompts up to 200,000 tokens; over that, input is $4 and output $18 per million.
- Gemini 3.1 Flash-Lite: Text, image and video input. Audio input costs more.
- Gemini 2.5 Flash-Lite: Text, image and video input. Audio input costs more.
DeepSeek
- Peak hours are 01:00 to 04:00 and 06:00 to 10:00 UTC, Monday to Friday. Everything else is off-peak, at half the price.
- No batch API.
- DeepSeek Flash: Peak prices shown; half that off-peak.
- DeepSeek V4 Pro: Peak prices shown; half that off-peak. No image input.
How it works
- 1
Describe a request
Type how many requests a month you make and the input and output tokens in one, or start from an example.
- 2
Add the discounts
Say how much of the input comes from the cache, and whether you can use the batch API or run DeepSeek off-peak.
- 3
Compare the cost
Every model is shown cheapest first, with its price per million tokens, per request and per month.
Questions
How is the cost worked out?
Each request costs its input tokens at the input price plus its output tokens at the output price, where prices are per million tokens. The part of the input served from the prompt cache is charged at the lower cached price. The monthly cost is that times your requests per month. Everything is list price in US dollars, before tax.
What is a token, and how many will I use?
A token is a piece of a word. In English, one token is roughly four characters, so 1,000 tokens is about 750 words. Providers count differently, so the same text can use a different number of tokens on each model. Look at the usage figures your provider returns with each response, or use its token counter, to get real numbers for your prompts.
What does prompt caching do?
When the start of your prompt is the same from one request to the next (a long system prompt, a document, a set of tools), the provider can serve it from a cache at a fraction of the normal input price. Type the share of your input you expect to come from the cache and the calculator prices that part at the cached rate. Anthropic also charges a one-off fee to write to its cache, which is not counted here.
What does the batch API change?
Anthropic, OpenAI and Google offer a batch API that takes your requests in bulk and returns the results within a day, for half the price. It suits work nobody is waiting for, such as classifying a backlog or summarising documents overnight. DeepSeek has no batch API.
Why is output so much dearer than input?
Generating text costs the provider more than reading it, so output tokens are priced at several times the input price. Models that think before they answer bill those thinking tokens as output too, so for those models the output figure should include the reasoning, not only the visible answer.
Do these prices change?
Often. They were checked on October 29, 2026 against each provider's pricing page, which is linked under the table. Some changes are announced ahead: Google has said Gemini 4 Argon doubles in price on 1 December 2026 and Gemini 3.8 Flash on 1 January 2027, and the calculator switches to the new price on those days. DeepSeek charges half price outside its peak hours, which you can switch on.
Is the cheapest model the best choice?
No. The cheapest model is only the cheapest per token. A smaller model that gets the task wrong, or needs several tries, can cost more than a better one that gets it right first time. Use this to find the models worth testing, then run your own prompts through two or three of them and compare the results as well as the bills.
Is anything I type sent to you?
No. Everything is worked out in your browser, and nothing you type is stored or sent to us.
More free tools
- Website AuditCheck a page for SEO, sharing, speed and technical problems and get a score with fixes.
- Link Preview CheckerSee how a link will look on Facebook, X, LinkedIn, Slack and WhatsApp, and fix what is missing.
- Dofollow Link CheckerSee every link on a page and whether it is dofollow, nofollow, sponsored or UGC, and check that a backlink counts.
- SERP Snippet PreviewPreview your title and meta description as search results show them, and see where they get cut.
- OG Image GeneratorMake a 1200 by 630 share image for your page in your browser, then download it as a PNG.
- OG Image GalleryBrowse share image designs and open any of them in the generator to make it your own.
- UTM Link BuilderBuild tracked links for every launch channel, with presets for Product Hunt, X, Reddit and newsletters.
- Robots.txt GeneratorMake a robots.txt with your sitemap, private paths and a switch for AI crawlers such as GPTBot and ClaudeBot.
- llms.txt GeneratorWrite the llms.txt file that tells ChatGPT, Claude and Perplexity what your site is and where its key pages are.
- PPP Pricing CalculatorSet fair prices by country: see a suggested discount and price for 190+ countries.
- Directory Badge GeneratorMake a Featured on badge with your text, icon or logo and colours, and download it as an SVG or PNG with the HTML to paste.
- Payment Fee CalculatorCompare what Stripe, PayPal, Paddle, Lemon Squeezy and Polar keep from every sale, or find the price to charge to keep a set amount.
One email a week
The AI products that launched this week and the top of last week's board. No ads in it, and you can leave with one click.
Launching a SaaS or AI product?
List it on Stackboo for free. It is ranked by votes, not budgets, and reviewed by a person.