Methodology

This page documents exactly how our estimates are produced so you can judge their reliability and reproduce them.

Token cost formula

cost = (input_tokens × input_price + output_tokens × output_price) ÷ 1,000,000

Prices are per million tokens. Cached input tokens are billed at the provider's cached rate; the batch share receives the provider's batch discount.

Token estimation

Where we estimate tokens from text, we use ~4 characters or ~0.75 words per token. Actual tokenisation varies by model tokenizer and language; code and non-English text use more tokens per word.

ROI formulas

monthly_savings = manual_cost − ai_assisted_cost
net_benefit     = monthly_savings − ai_operating_cost
ROI%            = net_benefit ÷ ai_operating_cost × 100
payback_months  = setup_cost ÷ net_benefit

Support automation

manual = tickets × handle_time_hrs × hourly_cost
ai_assisted = escalated × handle_time_hrs × hourly_cost
            + tickets × ai_cost_per_ticket + platform + QA

Assumptions

A safety margin (where offered) buffers usage spikes, taxes and estimation error.
Conversation history is approximated as half the conversation length on average.
Self-hosting capacity = GPUs × tokens/sec × seconds in month × utilisation.

Currency

All underlying figures are in USD. Other currencies use static, display-only conversion rates and may not reflect live FX. This is not a currency conversion service.

Taxes and discounts

Estimates exclude VAT/sales tax, committed-use or enterprise discounts, and free tiers unless you enter them. Your actual invoice may differ.

Limitations

These are planning estimates, not quotes. Provider pricing, rate limits, regional uplifts and model behaviour change over time. Always confirm against the official pricing pages on our data sources page.