TCO calculator

How much could you save?

Three questions about your AI usage, and an estimate of your yearly bill with and without TokenSaver.

1. How many AI requests per day?

A request is one call to the model: a question, a chat turn, an agent step.

2. How big is a typical request?

ShortChat, support, classification≈ 1,300tokens per request1,000 sent300 receivedMediumAssistant on your documents (RAG)≈ 5,500tokens per request5,000 sent500 receivedLongAgents and code, per step≈ 20,600tokens per request20,000 sent600 received

Typical values derived from official sources (listed under “How it is computed”). Adjust them in the advanced settings if you know yours.

3. Which model do you use most?

People using TokenSaver

Pro plan: €79 excl. VAT per person per month.

Advanced settings

For those who know their numbers. Everything else stays the same.

Tokens sent per request

Tokens received per request

Monthly growth (%)

Scenario bounds

	Prudent	Optimistic
Compression (%)
Cache hits (%)
Routed to cheaper model (%)

Cheaper modelEnterprise subscription (USD / month, from your quote)

Leave empty if you do not have a quote yet.

Default bounds are provisional and will be replaced by measured benchmarks.

How cautious should the estimate be?

Central

PrudentCentralOptimistic

With this setting, we assume that TokenSaver:

Your estimate, per year

Estimate

You could save

—/ year

Today, without TokenSaver

—

With TokenSaver

—

Yearly cost

Without TokenSaver

With TokenSaver · prudent to optimistic

Estimate. Results are for illustrative purposes only. Actual costs vary based on usage patterns, use case, number of users, infrastructure costs, and applicable discounts. This calculator is not a substitute for a formal cost analysis.

How it is computed

AI bill = requests × (tokens sent × input price + tokens received × output price), at each provider’s official list prices per 1M tokens.
Cache: requests answered from the cache do not reach the model.
Compression: the text sent with the remaining requests is shorter.
Routing: part of the remaining requests go to the cheaper model (Pro and Enterprise).
Subscription: Pro at €79 excl. VAT per person per month, converted to USD. Enterprise on quote.

Official list prices of each provider, standard tier, in USD per 1M tokens, as of October 2, 2026. Excludes negotiated discounts, batch pricing, provider prompt caching and taxes. Gemini 3.8 Flash: introductory price until 31 Dec 2026. DeepSeek: off-peak rates, peak hours cost double. € converted at 1 € = 1.15 $ (provisional rate).

Anthropic OpenAI Google Mistral DeepSeek

Sources for the request sizes

OpenRouter and a16z, State of AI, empirical study of 100 trillion tokens (arXiv): about 6K input and 400 output tokens per request on average in late 2025; programming requests routinely above 20K input tokens. arxiv.org
Microsoft Learn, Azure AI Search: recommended RAG chunk size of 512 tokens, the basis of the Medium estimate (about 8 retrieved chunks plus the prompt). learn.microsoft.com
OpenAI Help Center, What are tokens: 1 paragraph is about 100 tokens, 100 tokens about 75 words, the basis of the Short estimate. help.openai.com
Anthropic Engineering, multi-agent research system: agents use about 4 times the tokens of chat interactions, multi-agent systems about 15 times. www.anthropic.com

Measure it on your own traffic.

The 30-day hosted trial measures cache, compression and routing on your real requests, run by run.

Start the free trial Talk to us about Enterprise