LLM API Cost Calculator: Claude, GPT & Gemini

What Is LLM API Cost?

LLM API cost is the amount you pay a provider such as OpenAI, Anthropic or Google each time your software sends information to an AI model and receives a response.

Unlike a normal software subscription with one predictable monthly fee, API billing usually depends on usage. The more requests your application makes, the more text you send, and the more text the model generates, the more you are likely to pay.

Most text-model pricing is expressed as a price per one million tokens. Tokens are small pieces of text that models process rather than whole words. Your prompt consumes input tokens, while the model’s response consumes output tokens.

That simple formula becomes more complicated once prompt caching, long-context pricing, Batch APIs and different model tiers are involved. That is why the calculator above asks for the shape of your actual workload instead of comparing only the headline price of each model.

What Actually Determines Your AI API Bill?

Four factors usually have the largest effect on your monthly cost. Changing just one of them can completely change which model looks affordable.

Reusable Prompt Tokens vs. New Input Tokens

Not every part of an API request changes every time.

How Does Prompt Caching Affect LLM API Cost?

Prompt caching can reduce the cost of repeatedly sending the same large block of text to a model.

Imagine that your application sends identical instructions, tool definitions and reference material with almost every request. Without caching, that repeated context may be processed and billed at the normal input rate each time.

With caching, supported providers can charge a lower rate when that repeated portion is successfully reused. The exact billing rules differ between providers, which is why the calculator does not apply one universal caching discount.

What Does Cache Hit Rate Mean?

Cache hit rate estimates how often your reusable prompt actually benefits from cached pricing.

If you enter a 90% cache hit rate, the calculator assumes roughly nine out of every ten eligible requests successfully reuse the cached portion. The remaining requests are treated as cache misses and billed according to the provider’s relevant miss or normal-input pricing.

A high theoretical cache discount is therefore much less useful if your application rarely achieves a cache hit.

When Should You Use Batch API Pricing?

Batch APIs are designed for workloads that can be processed asynchronously rather than returning an answer immediately.

Turning on Batch API in the calculator lets you compare the lower asynchronous processing rates supported by each provider.

Why Do Output Tokens Matter So Much?

Developers often spend most of their time optimizing prompts, but generated output can be an even bigger part of the bill.

If your application routinely asks for long reports, code, detailed reasoning or multi-section responses, output tokens can accumulate quickly. Reasoning models can also consume billed output or reasoning tokens that are not always obvious from the visible length of the final answer.

Why Can Similar AI Models Have Very Different Costs?

AI providers usually offer several model tiers because different workloads need different levels of capability.

A lightweight model may be ideal for extraction, categorization, short answers or repetitive structured tasks. More capable models may justify a higher price for complex reasoning, difficult coding, long-context analysis or work where a wrong answer is expensive.

That means the cheapest model in your calculator result should be treated as the lowest-cost candidate, not automatically the winning model.

How Do You Estimate Monthly LLM API Cost?

Suppose your application makes 1,000 requests per day for 30 days.

Each request contains 4,000 reusable prompt tokens, 500 new input tokens and generates 600 output tokens.

The calculator applies each model’s input, output, caching, Batch and applicable long-context rules to that same workload, then ranks the resulting monthly estimates.

That makes it much more useful than comparing provider price tables manually, because the same application can produce very different monthly bills depending on the model and pricing features you use.

Why Won’t Token Counts Be Identical Across Claude, GPT and Gemini?

Each model family can tokenize the same text differently.

Code, punctuation, languages, formatting and ordinary prose may therefore produce slightly different token counts depending on the provider and tokenizer.

For early planning, using one representative token estimate is usually enough to compare cost ranges. For production budgeting, use actual API usage data or the relevant provider’s tokenizer against your real prompts.

What Costs Are Not Included?

The calculator focuses on normal text-model token costs. Your actual AI bill may contain additional charges depending on the provider and product features you use.

Which LLM Should You Choose?

Use cost to narrow the field, then test your finalists using the exact task your application needs to perform.

A useful evaluation might run the same 50 to 100 representative prompts through two or three candidate models and compare accuracy, consistency, latency, instruction following and failure rate.

A model that costs twice as much per token may still be the better business choice if it produces far fewer errors. The opposite is also true: there is little reason to pay flagship-model prices for a task a lightweight model already handles reliably.