Input vs output token spend
Output tokens usually cost several times more than input, so answer length is what really drives your bill.
Home / Dev & AI / AI & LLM Cost / AI Inference Cost Calculator
AI & LLM Cost
AI Inference Cost Calculator
Price model serving in production. Enter monthly inferences and token sizes, choose a model, and see the cost per inference, per 1,000 and per month with a live 2026 comparison.
Estimated monthly cost
$0
Cost per inference—
Cost per 1,000 inferences—
Input token cost—
Output token cost—
Tokens processed / month—
Annual estimate—
The same workload priced across AI models
Your monthly token volume run through popular 2026 models, cheapest first.
What the AI Inference Cost Calculator does
Inference is the cost of running a trained model in production, once per request your users make. Unlike training, which is a one-time capital cost, inference is a recurring operating cost that scales directly with traffic. On a hosted API it is billed per token, so cost per inference depends on how many input and output tokens each call uses.
This AI inference cost calculator prices your serving workload by multiplying inferences by tokens and applying the 2026 rate for your model. It reports a clean cost per inference and per thousand, the figures you need to price a feature or check unit economics, and compares models so you can serve at the lowest viable cost.
Heads up on pricing. The rates built into this tool are public list estimates for July 2026 and can change without notice. Providers also offer batch discounts, prompt caching and volume tiers that lower real costs, so treat the output as a planning estimate, not a quote.
The formula
cost per inference = (input tokens ÷ 1M × input rate) + (output tokens ÷ 1M × output rate)
monthly cost = cost per inference × monthly inferences
monthly cost = cost per inference × monthly inferences
Worked example
Serving 1,000,000 inferences a month at 1,000 input and 500 output tokens on GPT-5 mini costs 1,000M × $0.25 + 500M × $2 per million = $250 + $1,000 = about $1,250 a month, or $0.00125 per inference.
AI model pricing used in this calculator (July 2026 estimates)
| Model | Provider | Input / 1M tokens | Output / 1M tokens |
|---|---|---|---|
| GPT-5 | OpenAI | $1.25 | $10.00 |
| GPT-5 mini | OpenAI | $0.25 | $2.00 |
| Claude Opus 4.1 | Anthropic | $15.00 | $75.00 |
| Claude Sonnet 4.5 | Anthropic | $3.00 | $15.00 |
| Claude Haiku 4.5 | Anthropic | $1.00 | $5.00 |
| Gemini 2.5 Pro | $1.25 | $10.00 | |
| Gemini 2.5 Flash | $0.30 | $2.50 | |
| DeepSeek V3 | DeepSeek | $0.27 | $1.10 |
| Grok 4 | xAI | $3.00 | $15.00 |
| Llama 4 Maverick | Meta | $0.20 | $0.60 |
Prices are public list estimates for planning as of July 2026 and change often. Providers bill per token, where roughly 1,000 tokens equals about 750 words. Always confirm live rates on the provider pricing page before you set a budget.
Ways to lower your AI costs
- Match the model to the task. Route simple requests to a small model like GPT-5 mini, Gemini Flash or DeepSeek V3, and keep premium models for complex reasoning where quality pays off.
- Cache repeated context. Prompt caching means you stop paying full price for the same system prompt or documents on every call.
- Cap output length. Output tokens cost several times more than input, so limiting response length is often the fastest saving you can make.
- Trim prompts and history. Summarize old conversation turns and drop context the model does not actually need to answer well.
- Batch non-urgent work. Many providers offer batch pricing at around half the standard rate for jobs that can wait a few hours.
Frequently asked questions
What is the difference between training and inference cost?
Training is a one-time cost to build the model. Inference is the ongoing cost to run it for each user request. For most products, inference is the larger lifetime expense.
How do I get cost per inference?
Multiply your average input and output tokens by the model rates and add them. This tool does it for you and shows per inference and per thousand figures.
Is self-hosting cheaper for inference?
It can be at very high, steady volume, where renting GPUs and serving an open model beats per token API pricing. At low or spiky volume, hosted APIs are usually cheaper and simpler.
How do I cut inference costs?
Use a smaller model, shorten prompts and outputs, cache repeated context, and batch requests where latency allows.
Are these figures current?
They are July 2026 estimates. Confirm live rates and any batch discounts with the provider before pricing a feature.