MILK Bench Animas AI

OpenAI Codex · GPT-5.6 family · frozen prompt

Seventeen product launches.
One carton of milk.

Each cell is a greenfield one-shot of the same MILK / 01 brief. Only the model and reasoning level change. Results below include wall time, token totals from Codex logs, and an illustrative OpenAI API cost if the work had been billed as standard API traffic.

Signals

High-res interactive charts — hover for details, click a bar to open that site, filter by model, expand any chart fullscreen.

Duration by run
Tokens by run
Est. API cost by run
Cost share by model
Payload size by run

Showing all 17 runs · click any bar to open that site

Extremes

Click any row to open the site.

All runs

Click a column header to sort