共有:

Enterprise GA

ChatGPT Agent Enters
The Enterprise Ring.

ChatGPT Agent (formerly Operator) is now generally available on Enterprise and Edu plans. At the same time, the previously free Workspace Agents shift to credit-based billing starting July 6. The move from pilot to production, and from experiment to procurement, is happening in the same week.

AI Navigate Editorial2026.07.046 min read

Plus Pro Team Enterprise Edu GA Credits 7/6 → Billing starts
Enterprise and Edu now sit at the top of the plan staircase, while the credit-billing meter on the right begins ticking on July 6.
01

Where things stood

Operator was for the curious, not the CIO

Operator, the predecessor to ChatGPT Agent, launched in early 2025 as an experimental feature aimed at individual subscribers. It opened browsers, filled out forms, booked appointments, and wrote code — fascinating to watch, but hard for enterprises to actually adopt.

The reason was simple: there was no dedicated tier for enterprise customers to run Agent in production. The only options were Plus, Pro, and Team — plans designed for individuals and small teams — and the surrounding machinery that large organizations require (SSO, audit logs, DLP, data-retention policies) hadn't fully reached Agent's execution environment.

As a result, many enterprises wanted to try it, but stalled the moment procurement, IT, and legal sat at the same table. Pilots moved forward, but cross-department production adoption did not. That was the bottleneck.

Education institutions faced the same shape of problem. Rolling Agent out to students and researchers required integration with an existing Education contract, a unified admin console, and usage-based pricing that fit within academic budgets. Until those pieces were in place, embedding Agent in an accredited course was off the table.

02

Update — 2026-07-04

GA on Enterprise / Edu; Workspace Agents priced from 7/6

Plus Pro Team Enterprise Edu GA 2026-07-04 Workspace Agents Free thru 7/5 7/6 → Credits billed usage-based
Enterprise and Edu step onto the staircase officially, while Workspace Agents move from the free pool to a credit-metered, usage-based model.

On 2026-07-04, OpenAI announced that ChatGPT Agent is now generally available on Enterprise and Edu plans. Alongside the GA, they also confirmed that Workspace Agents move to credit-based billing on July 6. The window of free experimentation closes in just a few days.

Inside Enterprise and Edu, Agent execution is now wired into the same admin console customers already use for SSO, audit logs, and DLP. On the education side, the Edu plan integrates with existing Education contracts, making student-account distribution a realistic operational model rather than a one-off pilot.

GA
Enterprise / Edu
7/6
Credit billing begins
Workspace
Agents now paid
03

Before / After

From "try it out" to "budget for it"

The choices in plans and the shape of costs are being rewritten along two axes at once.

Before (Plus / Pro / Team)After — Enterprise / Edu GA
Pilots for individuals and small teamsTier designed for large-org and university rollout
Workspace Agents ran from a free poolCredit-based, usage-metered from July 6
SSO / audit logs / DLP were limitedWired into the existing Enterprise admin console
Schools tried it via ad-hoc team plansEdu integrates directly with Education contracts
Cost felt like "how much you used it"Cost accrues as concrete execution credits

The moment your organization crosses to the right column, your job shifts from "try it out" to "estimate it." How many people, how many runs, how many browser actions and tokens per month? Guesses that used to be tolerated now need to be actual numbers.

04

Rollout path

From provisioning accounts to running in production

With Enterprise / Edu GA now available, the internal rollout project largely falls into four stages.

01

Provision accounts, design permissions

Under the Enterprise or Edu contract, decide which departments and roles get Agent. Wire in SSO and audit logs, define DLP rules, and lock in the allowlist of external sites Agent may navigate — all before anyone starts real work.

02

Estimate credits, secure the budget

Using observed "credits per task" from prior pilots, forecast monthly consumption. Do not forget to include Workspace Agents usage from July 6 onward, which now hits the same credit meter.

03

Limited pilot

Run a one-month pilot with 2 to 3 departments and a few dozen users. Pull real credit consumption, success rates, and deviation cases from the logs, then reset the estimate to about 1.5x-2x of what you had assumed.

04

Production rollout

Roll out organization-wide. Schedule a quarterly usage review; where teams blow past expectations, revisit their ownership boundaries and operational rules rather than just paying the invoice.

05

Who feels it first

The teams that notice the shift immediately

Enterprise IT

Approval memos for "may we run Agent in production" can finally be written inside the Enterprise standard frame. SSO and audit logs dissolve much of the earlier hesitation — but building the first credible credit-budget estimate becomes a new part of the job.

Universities / Edu

With Edu folded into Education contracts, distributing to student accounts and allocating credits per faculty becomes an actual operating model. Individual research groups finally have a stable base to run agent experiments on.

Internal DevOps / platform

When Workspace Agents get a price tag, the internal "agent execution platform" gets a clearer identity. Usage metrics, chargeback across departments, and per-team dashboards move from nice-to-have to urgent.

Pilots are over.
It is procurement time.

06

Frontier

Agents move from "the lab" to "the SKU list"

Enterprise GA and the introduction of paid Workspace Agents are two sides of the same coin. One signals that large enterprises are finally entering the real adoption phase. The other signals that costs, previously tolerated on gut feel, now have to be added up as actual numbers.

Agent conversations have been happening on the floor since 2025. From here, they also happen in the conference room. "How much would this task cost per month if Agent handled it?" "Who is responsible when it makes a mistake?" "Where do the audit logs live?" — the language shifts from demo talk to procurement, legal, and IT talk.

Once that translation begins, the gap between "an organization that uses AI" and "an organization that doesn't" starts to appear plainly on org charts and budget tables. Companies that develop the muscle to put Agent through their approval process first will see Agent usage sitting on their quarterly reviews, then bound to KPIs the quarter after that.

The back half of 2026 looks likely to be when this quiet shift moves fastest. Pilots are over. It is procurement time.


At AI Navigate, we care less about the announcement itself and more about what starts to happen inside organizations because of it. When the language of the approval memo changes, so does the use of AI — slowly, but in a direction that does not reverse.