TL;DR

OpenAI's analytics guide describes how administrators can inspect activity, credits, tokens and task categories alongside engineering indicators. Those signals show usage; they do not by themselves prove productivity or return on investment. Pair spending with accepted outcomes and human review time. The guide's illustrative dashboards and hypothetical ROI calculation should not be treated as measured results for a typical team.

What changed

OpenAI published a guide to Admin Console analytics spanning ChatGPT Work and Codex. It describes views for active users, credits, tokens, task categories, and the share of merged code with Codex contributions. It also explains how admins can combine these signals with business data through the Admin API.

Usage is evidence of activity, not proof of productivity. OpenAI's examples include illustrative dashboard data and a hypothetical return-on-investment calculation. Those figures should not be read as measured results for a typical team. The post is a product guide, not a claim that every described feature launched that day.

What it means for your workflow

SaveMyToken editorial take

Track accepted work and review time alongside spending. More generated code or more tokens can increase activity without improving the outcome your team cares about.