LLM API Cost Calculator
Estimate your monthly Claude, GPT and Gemini API bill — with prompt caching, batch discounts and long-context pricing
| Model | Per request | Per month | Caching saves | |
|---|---|---|---|---|
Claude Haiku 5.5 Anthropic | $0.00049 | $43.74 | $28.26 | |
GPT-6 Luna OpenAI | $0.00049 | $43.74 | $28.26 | |
Gemini 3.5 Flash-Lite Google | $0.0018 | $165 | $87.48 | Implicit caching; no write fee |
Gemini 3.8 Flash Google | $0.0036 | $321 | $219 | From Jan 1, 2027: $643/month Implicit caching; no write fee |
GPT-6.1 Terra OpenAI | $0.0047 | $421 | $299 | |
Gemini 4 Argon Google | $0.0092 | $824 | $616 | From Dec 1, 2026: $1,649/month Implicit caching; no write fee |
Claude Sonnet 5.5 Anthropic | $0.0094 | $842 | $598 | |
GPT-6.1 Sol OpenAI | $0.0094 | $842 | $598 | |
Gemini 3.1 Pro (preview) Google | $0.010 | $929 | $583 | Implicit caching; no write fee |
Claude Opus 5.5 Anthropic | $0.019 | $1,685 | $1,195 | |
Claude Fable 5.1 Anthropic | $0.046 | $4,131 | $3,069 | |
GPT-6 Astra OpenAI | $0.049 | $4,374 | $2,826 |
Per request on Claude Haiku 5.5: input $0.00020, cache reads $0.000036, cache writes $0.000050, output $0.00020. A month is 30 days.
List prices from the Anthropic, OpenAI and Google pricing pages, checked Oct 31, 2026; priced for Oct 31, 2026. Cache writes use Anthropic's 5-minute and OpenAI's 30-minute rates. Taxes, fast/priority tiers, data-residency surcharges and Gemini's explicit-cache storage fees aren't included.
Not sure how many tokens your prompt is? Paste it into the AI Token Counter.
Was the LLM API Cost Calculator helpful?
About this tool
The LLM API Cost Calculator turns a workload into a monthly bill: how many requests you send, how many input and output tokens each one uses, how much of the prompt is a reusable prefix, how often that prefix is a cache hit, and how much traffic can go through the Batch API. It prices the same workload on current Claude, GPT and Gemini models side by side, so you can see what a model switch, a better cache hit rate or batching would save. Prices are the official list prices from Anthropic's, OpenAI's and Google's pricing pages (the date they were checked is shown under the table). It models each provider's own rules: Anthropic's cache-write premium, OpenAI's long-context surcharge above 272K input tokens, Claude Haiku 5.5's higher rate for prompts over 100K tokens, Gemini's implicit caching, and price changes Google has already announced, which switch on automatically on their start date. Everything is calculated in your browser; nothing you type is sent anywhere.
When to use it
- →Budgeting a new AI feature before launch
- →Comparing what the same workload costs on Claude, GPT and Gemini
- →Seeing how much prompt caching saves on a long system prompt or tool list
- →Deciding whether nightly jobs should move to the Batch API
- →Explaining an LLM bill to finance with per-request and monthly numbers
Tips
- ◆Count thinking / reasoning tokens as output — they're billed at the output rate even though you don't see them.
- ◆A cacheable prefix only helps if it's byte-for-byte identical on every request: put the static system prompt and tools first and anything that changes after them.
- ◆Paste a real prompt into the AI Token Counter to get a realistic input size.
- ◆Long-context surcharges apply to the whole request once the input crosses the threshold, not just the extra tokens.
Frequently asked questions
Where do the prices come from?
From each provider's official API pricing page, checked on the date shown under the results. They're list prices for standard requests; enterprise discounts, taxes and fast or priority tiers aren't included.
How is prompt caching priced?
A cache hit costs a fraction of the input price (for example 5% on Claude Opus 5.5). A miss pays to write the prefix to the cache — 1.25× the input price for Anthropic's 5-minute cache, and OpenAI's 30-minute write price. Gemini's implicit caching has no write fee.
What does the Batch API change?
Batch requests cost 50% of the standard price and complete within 24 hours. The slider sets the share of your traffic that can wait that long.
Why does a model get more expensive above a certain prompt size?
Some models charge more for long prompts: GPT-6 models above 272K input tokens, Gemini 3.1 Pro above 200K and Claude Haiku 5.5 above 100K. The higher price applies to the whole request.
ToolNinja