Why does the cheapest model change with my token ratio?

Providers price output tokens several times higher than input. A long-prompt, short-answer task favors models with cheap input, while a short-prompt, long-answer task favors cheap output. Changing the ratio re-orders the heatmap.

Are the prices current?

The model prices are editable presets based on published list prices and labelled as estimates. Providers change pricing often, so confirm the current rate in your provider dashboard before budgeting.

What do the colors mean?

Color encodes relative cost across the models shown — green is the cheapest, red the most expensive, with a smooth gradient between. It is relative to the current list, so filtering by tier rescales the colors.

Is my data sent anywhere?

No. The heatmap is computed entirely in your browser. Nothing you enter is uploaded, stored or logged.

Token Cost Heatmap by Model

Token cost heatmap by model

Picking the cheapest model is not just “smallest number wins” — it depends on your prompt-to-completion ratio, because providers price output tokens far higher than input. This heatmap costs your specific request across 30+ models and colors each row by price, so the best-value choice is obvious at a glance.

How it works

For every model the cost of one request is:

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

The tool computes this for each model at your token counts, sorts cheapest to most expensive, and maps cost onto a green-to-red color scale relative to the models shown. A quality-tier filter lets you compare flagship, mid-range and fast models on equal footing instead of mixing a frontier model with a budget one.

Tips for choosing a model

Match the ratio to the task. Summarizing a long document (big input, small output) rewards models with cheap input; brainstorming (small input, big output) rewards cheap output.
Start in the fast tier. For classification, extraction and routing, a fast/mini model is often 10–20× cheaper and good enough.
Reserve frontier models. Use them where reasoning quality clearly moves the needle, not as a default.
Re-run when prices change. Edit the presets to your current contracted rates for an accurate ranking.