OpenAI starts charging for Workspace Agents
ChatGPT Workspace Agents in Business, Enterprise, and Edu plans began drawing workspace credits on July 6, ending the free preview that launched in April 2026.
Workspace Agent runs inside ChatGPT started drawing workspace credits on July 6, 2026 for Business, Enterprise, and Edu plan customers. Before this date, Workspace Agent runs had been free since the feature launched on April 22, 2026.
Agent runs triggered outside ChatGPT, such as agents responding in Slack channels, remain in free preview as of this change.
How credits are consumed
Credits are metered on three token streams per agent run: input tokens, cached input tokens, and output tokens. OpenAI’s worked example uses 20,000 input tokens plus 80,000 cached input tokens plus 5,000 output tokens, which equals approximately 7.25 credits. A typical end-to-end GPT-5.5 agent run costs roughly 5 to 25 credits depending on task complexity and how much context the agent retrieves.
All three affected plan tiers can purchase additional workspace credits on top of their seat subscription.
Why this matters now
The April to July window was intentional: free preview runs gave enterprise customers time to evaluate whether Workspace Agents added value before billing started. For most teams, the first month of actual credit charges will be the first reliable signal of what their real usage pattern costs.
The transition to credit-based billing also creates a new optimization surface. Agents that retrieve large context windows cost more. Agents that cache frequently used context cost less. If your team has been running agents without thinking about token efficiency, that changes now.
ChatGPT Enterprise already has spend controls at the workspace level; the credit system integrates with those limits. The Fable 5 usage credits brief from last week covered Anthropic’s parallel move to a credit-based model for its own managed-agent tier.