LLM API Cost Calculator
Paste one representative prompt, set your output length and requests per day, and see what a month costs on 13 current GPT, Claude and Gemini models — verified prices, client-side.
0
GPT tokens (exact)
$0
Input cost, cheapest model
0
Words
0
Characters
Loading tokenizer (≈2 MB, once)…
Each coloured block is one GPT token (o200k). Long or unusual words split into several; common English words are usually one.
Does it fit the context window?
How much of each model's limit your prompt uses, leaving room for the output length set below.
What this prompt costs.
Per request, and per month at your volume.
| Model | Your tokens | Input $/1M | Output $/1M | Per request | Per month |
|---|
Prices per million tokens from the providers' official pricing pages, verified . Standard (≤200K-token) tier, no caching or batch discounts. GPT counts are exact (o200k tokenizer, used by GPT-4o and later); Claude and Gemini counts are estimates — those tokenizers aren't public. Promotional rates are marked ◦. Confirm current pricing with the provider before budgeting.
How to budget an LLM feature in four numbers
Every LLM bill reduces to input tokens × input price + output tokens × output price, summed over requests. The trap is that three of those four numbers are usually guessed. This calculator pins the first (tokenizing your real prompt), asks you for the second and the request count, and pulls the prices from a registry verified against the providers' own pages, so the only assumption left is yours.
Where the money goes. On current models output costs 5–6× input per token. A 1,200-token prompt with a 400-token answer on GPT-5.6 Terra is $0.0024 in, $0.0048 out — the reply is two-thirds of the bill. Cutting the answer from 400 to 200 tokens with a tighter instruction saves more than moving from Terra to Luna would on the input side. Check the ratio before you shop models.
Where the table is conservative. It ignores prompt caching (10% of input price on reused prefixes, on both OpenAI and Anthropic), batch discounts (50%, both), and free tiers. If your system prompt is large and constant, your real input cost may be well under the table. It also ignores long-context surcharges above 200K tokens, where real cost is higher.
A sanity check for the monthly figure. Multiply your requests per day by 30 and by the per-request cost, then add 10–20% for retries, fallbacks and the occasional long conversation. If that number is within budget on the model that meets your quality bar, ship; if not, the levers in order are: shorter output, cached system prompt, smaller model for the easy cases, then a cheaper provider.
How to use LLM API Cost Calculator.
- 1
Paste a prompt that represents a typical request — system message plus a realistic user turn
- 2
Enter the output tokens you usually get back (a short answer is ~100–300; a page of text ~500–700)
- 3
Enter requests per day for the feature you are budgeting
- 4
Read the per-month column; the cheapest model that fits your quality bar and context window is your answer
Why use this tool.
Answer "what will this feature cost us per month" in thirty seconds
Compare providers on your workload rather than their headline per-token rates
See the output-token multiplier — the part most cost estimates miss
Keep proprietary prompts private while you budget
About LLM API Cost Calculator.
Most LLM pricing comparisons list dollars per million tokens and stop there, which tells you nothing about what your feature will cost. This calculator starts from your actual prompt — tokenized in your browser — adds the reply length you expect and your daily request volume, and shows per-request and monthly cost for thirteen current OpenAI, Anthropic and Google models side by side, with prices verified against the official pages and promotional rates flagged. It surfaces the two effects that dominate real bills: output tokens priced at several times the input rate, and tokenizer differences that make a "cheaper" model bill more tokens. Nothing you paste is uploaded.
Frequently asked questions.
Why does output length matter so much?
Output tokens are billed at 5–6× the input rate on most current models (GPT-5.6 is exactly 6×). A prompt of 1,000 input tokens with a 500-token reply spends roughly three-quarters of its cost on the reply. Getting the model to answer concisely is often a bigger saving than switching models.
Are caching and batch discounts included?
No — the table shows standard, uncached, real-time prices so it is comparable across providers. OpenAI and Anthropic bill cached prompt prefixes at 10% (or less) of the input rate, and both offer ~50% off for batch (asynchronous) processing. If your workload qualifies, treat the table as the upper bound.
What do the ~ marks and ◦ marks mean?
~ marks an estimated token count (Claude and Gemini tokenizers are not public; GPT counts are exact). ◦ marks a promotional price the provider has said will change — GPT-5.6 Sol's rate runs at least through 21 November 2026 and Gemini 3.8 Flash's through 31 December 2026.
How often are prices updated?
The registry records the date it was last verified against each provider's official pricing page (shown under the table). Prices change several times a year; confirm before committing a budget.
Can I model long-context pricing?
Not yet — the table uses the standard ≤200K tier. Gemini Pro and OpenAI's newest models charge more above 200K tokens (roughly double on input). If your prompt bar shows you above 200K, expect a higher rate than the table.
Guides for this tool.
- LLM API pricing compared (September 2026): what a real request costs on GPT, Claude and Gemini
Thirteen current OpenAI, Anthropic and Google models priced per million tokens and, more usefully, per request and per month. Verified against official pricing pages, with the caching, batch and long-context rules that change the answer.