As of today, June 10, that leaves administrators exactly 26 days to figure out what their agent fleets will cost in credits, set spend controls, and decide which automations survive once the meter starts running.
The stakes are concrete because the workloads are. Companies have used the free window to hand agents real jobs: qualifying sales leads, preparing month-end accounting workpapers, triaging IT requests, and answering employee questions in Slack. Every one of those workflows acquires a per-run price on July 6, and OpenAI publishes the consumption rates but not a public dollar figure per credit, which makes the next 26 days a budgeting exercise with one variable only your account team can fill in.
What Changes for ChatGPT Workspace Agents on July 6, 2026?
Workspace Agents launched on April 22 in research preview for ChatGPT Business, Enterprise, Edu and Teachers plans, and OpenAI's announcement post initially promised free usage only until May 6, with credit pricing starting that day. Decrypt reported the original May 6 cutoff at launch.
That deadline moved. On May 22, OpenAI declared workspace agents generally available across Business, Enterprise and Edu plans, added admin-console visibility into agent activity, and in the same release note extended the free period to July 6, 2026. Six days later it added GPT-5.5 support with adjustable reasoning effort, role-based publishing permissions, speech output, and smarter Slack thread replies. The two-month extension bought customers time, but it also bought OpenAI a larger installed base of shared agents that teams now depend on, exactly the position a vendor wants its users in when usage-based billing begins.
From Custom GPTs to Codex Agents Running in the Cloud
Workspace Agents are not a rebrand of custom GPTs; they are an architectural replacement. OpenAI describes them as "an evolution of GPTs" powered by Codex, running in the cloud rather than inside a chat session, which is what lets them keep working after the user logs off, run on schedules, retain memory across projects, and execute multi-step workflows that write or run code and call connected apps. VentureBeat framed them as the successor to custom GPTs that plugs directly into Slack, Salesforce and other enterprise systems, and OpenAI says teams can interact with agents in ChatGPT and Slack today with more surfaces coming.
The free preview did what free previews are designed to do: it normalized delegation. Rippling AI engineering lead Ankur Bhatt said in OpenAI's launch post that a sales-opportunity agent built without an engineering team "researches accounts, summarizes Gong calls, and posts deal briefs directly into the team's Slack room," replacing five to six hours of weekly rep work per deal. OpenAI's own accounting team runs an agent that prepares journal entries, balance-sheet reconciliations and variance analysis for month-end close. Those are precisely the always-on, scheduled, shared workloads that token metering will price.
How Are Workspace Agent Credits Calculated?
The mechanics live in OpenAI's ChatGPT rate card. Workspace Agent runs have no fixed per-run price. Instead, each run is metered across three token streams: credits per million input tokens, credits per million cached input tokens, and credits per million output tokens. The final cost varies with task complexity, input size, cache hits and output length.
OpenAI's worked example makes the arithmetic tangible: a GPT-5.5 Workspace Agent run consuming 20,000 input tokens, 80,000 cached input tokens and 5,000 output tokens costs about 7.25 credits. Cached input is the discount lane, billed at 12.50 credits per million tokens for GPT-5.5, so the 80,000 cached tokens in that example contribute exactly 1 credit (80,000 divided by 1,000,000, multiplied by 12.50). Cached tokens are cheaper because the model reuses context it has already processed, instructions, files and conversation history that have not changed between turns, rather than recomputing it. The rate card says a typical end-to-end GPT-5.5 agent run lands between 5 and 25 credits, a fivefold spread driven mostly by how much fresh context the agent ingests and how much it writes.
The design consequence matters for anyone building agents this month: an agent with stable instructions and a stable knowledge base rides the cached-input rate on every scheduled run, while an agent that re-ingests large fresh documents each time pays full input price repeatedly. The same business workflow can sit at either end of that 5-to-25-credit range depending on how it is architected.
One more free lunch survives: OpenAI's flexible pricing FAQ states that when ChatGPT automatically routes a request to a mini model, zero credits are charged, because lightweight tasks never touch the full-size model.
Slack Runs Stay Free: The Carve-Out Admins Should Map First
The July 6 metering applies to Workspace Agent runs invoked within ChatGPT. Per the rate card, runs invoked outside ChatGPT, such as agents responding in connected Slack channels, continue in free preview past the cutover. OpenAI has not published an end date for that carve-out.
That split creates a short-term arbitrage and a planning trap. Teams whose agents primarily answer questions and pick up requests in Slack pay nothing for now. Teams whose agents run on schedules inside ChatGPT, the weekly metrics reporters and month-end close preparers, start consuming credits on day one. Admins should map every agent by invocation surface before assuming a budget, and should also assume the Slack preview eventually ends on its own date.
What Should Admins Do Before July 6?
The dollar question comes first, and OpenAI does not answer it publicly. The flexible pricing FAQ explains that Business workspace owners buy credit packs through Settings and Billing, with credits pooled across all users and valid for 12 months, while Enterprise and Edu customers purchase a shared credit pool through their OpenAI account team as part of their contract, with allocation and expiration set in the order form. Translating 7.25 credits into dollars therefore requires knowing what your organization pays per credit under its own plan, which is the first call to make this week.
The control toolkit already exists. Business admins can configure usage alerts and automatic recharge, or leave recharge off so advanced features pause when the pool empties and users send in-product requests for more. Enterprise and Edu owners can set usage alerts and hard overage limits (a limit of zero blocks overages entirely) and use role-based access controls to set spend rules by group. The admin console shows each agent's total runs, unique users and usage over time, and OpenAI's Compliance API exposes every agent's configuration, updates and run history, the raw material for a cost-per-workflow estimate.
The triage logic follows from the metering math: inventory every published agent, rank scheduled in-ChatGPT runs by frequency and context size, restructure heavy agents to maximize cached input, route what can live in Slack to Slack, and retire the experiments nobody owns. A daily agent averaging 25 credits per run compounds to roughly 9,000 credits a year by itself; multiplied across dozens of shared agents, that is the spreadsheet OpenAI's 26-day runway exists for. Usage-based pricing rewards organizations that measure and quietly penalizes the ones that do not, and the free measuring window closes on July 6.
Frequently Asked Questions
When does OpenAI start charging for Workspace Agents?
Credit-based pricing for ChatGPT Workspace Agents begins on July 6, 2026, when the extended free period ends. The original free window was set to expire May 6 before OpenAI pushed it back in a May 22 release note.
How many credits does a Workspace Agent run cost?
OpenAI's rate card says a typical end-to-end run on GPT-5.5 consumes between 5 and 25 credits, with no fixed per-run price. A worked example with 20,000 input tokens, 80,000 cached input tokens and 5,000 output tokens comes to about 7.25 credits.
Are Workspace Agents in Slack also being charged?
Not yet. The July 6 rates apply to agent runs invoked within ChatGPT, while runs invoked outside ChatGPT, such as agents responding in Slack channels, remain in free preview. OpenAI has not announced an end date for that preview.
What are ChatGPT Workspace Agents?
They are Codex-powered shared agents introduced on April 22, 2026 as the successor to custom GPTs, available on ChatGPT Business, Enterprise, Edu and Teachers plans. They run in the cloud, work on schedules, keep memory across projects, and connect to tools like Slack and Salesforce to complete multi-step workflows.
ⓒ 2026 TECHTIMES.com All rights reserved. Do not reproduce without permission.
