Agent Credits: Definition
Agent Credits are OpenAI's usage-based billing unit for Workspace Agents, consumed per agent invocation based on model, runtime length, and tool calls.
Longer definition
When the Workspace Agents research preview ended on May 6, 2026, OpenAI shifted to credit-based billing at the workspace level. Each agent invocation consumes credits, and the exact amount depends on which model is used, how long the agent runs, how many tool calls it makes, and how much data it processes. Credits are purchased in bundles at the workspace level, with overage alerts and per-agent caps configurable by admins.
In practice
Single-run business agents (Weekly Metrics Reporter, Invoice Reviewer, Meeting Prep) consume very few credits per run. Chatty agents fired hundreds of times per day (Lead Outreach at scale, Support Triage on high-volume channels) consume more. Sized correctly, most deployments settle into a predictable monthly amount per active agent, and high-volume setups with many agents scale from there.
Common mistakes
- Not setting per-agent credit caps, since a mis-configured agent can consume unexpected credits
- Comparing credit cost in isolation. The real comparison is credits versus human hours saved, not credits versus nothing
- Skipping credit estimation during build. Size agents for a predictable cost rather than discovering it at end-of-month
See also
Want this shipped for your team?
20 minutes on video. We'll tell you if it's a fit and what it would take.