AI Cost Calculator: Token & LLM Price Comparison

Estimate token and API costs for GPT, Claude, Gemini, DeepSeek and more, and compare models on the same usage. Multilingual, no sign-up.

ModelProvider$/1M in$/1M outPer requestTotal
Gemini 2.5 Flash-LiteCheapestGoogle$0.1$0.4$0.0003$0.3
Claude Haiku 5.5Anthropic$0.1$0.5$0.00035$0.35
GPT-6 LunaOpenAI$0.1$0.5$0.00035$0.35
GPT-4o miniOpenAI$0.15$0.6$0.00045$0.45
DeepSeek Flash (off-peak)DeepSeek$0.15$0.6$0.00045$0.45
GPT-5 miniOpenAI$0.25$2$0.00125$1.25
Gemini 2.5 FlashGoogle$0.3$2.5$0.00155$1.55
DeepSeek V4 Pro (off-peak)DeepSeek$0.66$1.98$0.00165$1.65
Gemini 2.5 Pro (≤200k)Google$1.25$10$0.00625$6.25
Claude Sonnet 5.5Anthropic$2$10$0.007$7
GPT-6.1 SolOpenAI$2$10$0.007$7
Gemini 4 Argon (promo)Google$2$10$0.007$7
GPT-4oOpenAI$2.5$10$0.0075$7.5
Gemini 3.1 Pro (≤200k)Google$2$12$0.008$8
Claude Opus 5.5Anthropic$4$20$0.014$14
GPT-5.6 SolOpenAI$4$20$0.014$14
GPT-5.5OpenAI$5$30$0.02$20
Claude Fable 5.1Anthropic$10$50$0.035$35
GPT-5.5 ProOpenAI$30$180$0.12$120

Prices are indicative, last updated — verify on the provider's official page before relying on them.

Bulk calculation (CSV)

Upload a CSV with columns label, input_tokens, output_tokens, requests; it computes each row's cost on the model you pick and lets you download the results.

Share this calculation

The link keeps basic share tracking.

What does this tool do?

Enter your input and output token counts and a request count; the tool computes the per-request and total API cost for each model and sorts from cheapest to most expensive.

You can also paste your prompt text to estimate tokens. The goal is to compare, in seconds, how much each model would cost for a given project.

Why do input and output prices differ?

LLM providers usually price input tokens below output tokens, because generating output costs more. Long answers (lots of output) drive the cost up quickly.

Some providers offer much cheaper cached input, which saves a lot on repeated system prompts.

How do I reduce API cost?

Pick a smaller/cheaper model, shorten prompts, drop unneeded context, cap the output length (max tokens), and use prompt caching and batch discounts where available.

The most practical approach is to try the same task on a few models and compare the quality/cost trade-off using this table.

How are tokens estimated?

Real token counts are model-specific; this tool uses the common English approximation of ~4 characters per token. Non-Latin scripts usually use more tokens, so treat the estimate as a planning figure.

The goal is not only to produce a number, but to make the result understandable for practical decisions.

Limitations and official sources

Prices change often and differ by provider, so this table is indicative and dated. For a real cost decision, always verify on the provider's official pricing page (sources below).

The goal is not only to produce a number, but to make the result understandable for practical decisions.

Why this is needed and known limitations

This kind of tool is useful for budgeting, quoting, pre-analysis, and communication clarity.

It should not replace professional review or official compliance checks for final high-impact decisions.

Was this tool helpful?