OpenAI Pushes Workspace Agent Billing to July 6, 2026 — Here's What to Budget For
OpenAI has extended free access to ChatGPT workspace agents until July 6, 2026, giving Enterprise and Edu admins more time to plan credit-based budgets.
Update, 6 July 2026: ChatGPT Workspace Agent credit billing goes live today — free period ends
The extended free period described in the original article is now over. Credit-based pricing for ChatGPT Workspace Agents takes effect today across Business, Enterprise, and Edu plans.
Costs are not fixed per run. Usage depends on input tokens, cached input tokens, and output length. OpenAI’s rate card puts a typical GPT-5.5 agent run at 5 to 25 credits; a worked example of 20,000 input tokens, 80,000 cached input tokens, and 5,000 output tokens comes to roughly 7.25 credits.
One exception: agents invoked outside ChatGPT, such as those responding in Slack channels, remain in free preview with no end date announced.
Admins need to act today. Business workspace owners can set usage alerts under Settings > Billing > Usage Alerts. Enterprise and Edu admins have two separate controls: a per-user monthly usage limit and a workspace overage limit. The overage limit defaults to “No limit,” meaning uncapped charges apply immediately if it has not been configured. The usage limits guide explains how workspace, group, and user overrides interact. If a workspace exhausts its credit pool, credit-dependent features including Codex become unavailable until credits are added.
If your team has been running ChatGPT workspace agents without watching the clock, you now have a firm date to put in the diary: July 6, 2026 is when credit-based billing kicks in for workspace agent runs on ChatGPT Business, Enterprise, and Edu plans.
OpenAI originally launched workspace agents in research preview on April 22, promising free usage only through May 6. When the feature went generally available on May 22, that deadline quietly extended to July 6. Two extra months is a genuine gift for planning, but that window is closing, and the billing mechanics are worth understanding before you hit it.
What workspace agents actually do
Workspace agents are OpenAI’s answer to the question “what if a custom GPT could take actions, not just answer questions?” Built on the Codex model, they can own multi-step workflows, connect to tools like Slack, GitHub, Snowflake, and Databricks, follow team processes, and be shared across a workspace so one person builds and everyone else uses. Think of them less as a chatbot and more as a colleague you assign recurring tasks to.
Since general availability, OpenAI has added GPT-5.5 support with adjustable reasoning effort, role-based publishing permissions, speech output, and smarter Slack thread replies. Admins now have visibility into agent activity directly from the admin console, including Agent ID, connected apps, memory files, schedules, and usage analytics like unique users and runs over time.
How the credit pricing actually works
This is where it gets specific, and specific is useful here.
Workspace agent runs are priced on token consumption, not a flat per-run fee. OpenAI’s rate card puts a typical end-to-end run somewhere between 5 and 25 credits. As a worked example: a run using 20,000 input tokens, 80,000 cached input tokens, and 5,000 output tokens comes to roughly 7.25 credits. Cached tokens cost significantly less, so agents that repeatedly work with the same context (a shared document, a standard template) will be cheaper to run than agents processing fresh input each time.
There is no fixed credit cost per run. The final figure depends on task complexity, input size, how much of that input is cached, and output length. This means a quick agent run that summarises a short Slack thread and a full research-and-draft workflow are priced very differently, as you would expect.
One important nuance: runs invoked within ChatGPT are what become billable on July 6. Runs triggered externally, such as from within Slack, remain in free preview until June 2026, so that timeline has already passed for external integrations by the time you read this.
What admins should do before July 6
The practical to-do list is short but worth completing properly.
Set up usage alerts now. Business workspace owners can configure credit usage alerts in Workspace settings under Billing. Enterprise and Edu admins get both usage alerts and hard overage limits, which means you can cap spend rather than just monitor it. If a workspace runs out of credits for a usage-based feature, that feature becomes unavailable until credits are added, so hard limits are a genuine control mechanism, not just a reporting tool.
Review which agents are running and how often. The admin console now shows agent activity including run counts and unique users over time. Before billing starts, this is the right moment to identify which agents are genuinely embedded in team workflows versus ones that were built during the preview and quietly abandoned.
Check your credit balance and automatic reload settings. Admins can configure a minimum balance threshold so ChatGPT automatically charges a payment method when credits drop below that level, topping up to a target balance. This prevents workflow interruptions but also means spend can continue beyond what you might expect if you are not monitoring it actively.
Audit agent permissions while you are in there. Agent builders can set safeguards on which actions agents can take for each connected app. General availability is a good moment to tighten those if they were left broad during the preview period.
For education plans specifically
Verified US K-12 educators on the ChatGPT for Teachers plan have free access through June 2028, with workspace agents included in that offer at no additional cost during the preview period. Edu plan admins do get credit-based billing from July 6, but the controls available to them, including hard overage limits, are the same as Enterprise. Plugin sharing is also enabled by default for Edu accounts, whereas Enterprise accounts have it off by default.
The broader picture
The two-month extension from May to July gave customers time to embed agents into real workflows. It also, not coincidentally, gave OpenAI time to build the kind of team dependency that makes usage-based billing a natural next step rather than a surprise charge. That is a reasonable observation to hold in mind when planning budgets: the agents that have become genuinely useful are exactly the ones that will generate the most credits once billing starts.
The token-based pricing model does reward efficient agent design. Agents that work with cached inputs, produce concise outputs, and avoid unnecessary tool calls will cost less to run. If you have agents that were built quickly during the preview, it is worth a quick review to see if they can be made leaner before the meter starts.
July 6 is soon. The admin console work described above takes an afternoon at most, and it is considerably easier to do it now than to explain a large unexpected credit bill in Q3.