LLM Cost Calculator
Estimate the monthly cost of any OpenAI, Anthropic, or Google model, then see how much caching could save you.
Your workload
Share of requests served from cache instead of the model. Repetitive workloads routinely hit 40 to 70 percent.
Per request
--
Per year
--
Every model, your workload
1,500 in / 500 out tokens × 100,000 req. Cheapest first. Tap a row to select.
List prices per 1M tokens, standard tier. Estimates only. Confirm current rates with each provider.
How LLM pricing works
LLM providers bill per token, roughly 4 characters or about 0.75 words of English. Every request is priced in two parts: the input tokens you send (your prompt, system message, and any retrieved context) and the output tokens the model generates. Output is almost always the more expensive of the two.
Your monthly bill is the per-request cost multiplied by your request volume. The calculator prices your specific workload across every model in our database, so you can compare real cost instead of a generic benchmark.
How to cut your LLM costs
- Cache repeated calls. Many production workloads send the same or near-identical prompts repeatedly. Serving those from a cache removes the model call entirely.
- Right-size the model. A smaller model often handles routine tasks at a fraction of the cost. Use the table above to see the gap.
- Trim the prompt. Shorter system messages and tighter retrieved context cut input tokens on every call.
- Cap output length. Set a max output token limit so the model stops once it has answered.
Frequently asked questions
How is LLM cost calculated?
Cost per request = (input tokens / 1,000,000 x input price) + (output tokens / 1,000,000 x output price). Multiply by your monthly request volume for the monthly cost. Input and output tokens are priced separately, and output is usually more expensive.
What is a token?
A token is a chunk of text a model reads or writes, roughly 4 characters or 0.75 English words. Both your prompt (input) and the model's response (output) are billed in tokens.
Which LLM is cheapest?
It depends on your input-to-output ratio, but lightweight models like GPT-4o mini, Claude Haiku, and Gemini Flash are the cheapest per token, often 10 to 20 times less than frontier models. The comparison table ranks every model for your exact workload.
How much can response caching save?
A cache hit serves a repeated prompt from storage instead of calling the model, so that request costs effectively nothing. At a 60% cache hit rate you cut model spend by about 60%. Fluiq's optimize() builds a cache profile from your real traces and serves duplicates automatically.
Where do these prices come from?
Prices are list rates per 1M tokens for the standard tier, served from Fluiq's pricing database. Providers change pricing often, so treat the result as a planning estimate, confirm with the provider, and use the report link to flag anything out of date.
Stop estimating. Start measuring.
Fluiq traces every LLM call with real token counts and USD cost at provider rates, then caches repeated prompts automatically. Two lines of Python.