OpenAI Ends Free Trial for ChatGPT Business Agents, Shifts to Credit Billing

OpenAI has ended the free trial period for workspace agents in ChatGPT. Starting July 6, 2026, every run of such an agent on Business, Enterprise, Edu, and Teachers plans will incur a cost, billed under a credit system tied to token usage rather than a flat per-run rate.

Workspace agents are a feature OpenAI introduced in spring 2026, built on the Codex model, letting teams create shared assistants for repetitive office tasks such as qualifying sales leads, preparing month-end accounting documents, handling IT helpdesk tickets, or answering employee questions on Slack. Once built by one person, an agent can be used and further developed by the whole team.

How the new billing works

During the trial period, running from the feature's launch through July 6, companies could run agents at no cost, which let many organizations roll them into daily workflows without cost pressure. Now every agent run draws from a monthly credit pool assigned to the account, with the number of credits consumed depending on the model and the length of the task, counted as input tokens, cached tokens, and output tokens.

OpenAI has published a table of credit consumption but has not disclosed a public dollar rate for a single credit, meaning companies cannot convert their usage into a budget on their own without contacting their account manager. That turns expense planning overnight from a simple flat rate into a variable calculation that depends on how intensively individual teams use the agents.

Exception for external integrations

The new rates apply only to runs triggered directly inside the ChatGPT app. Agents operating outside ChatGPT, for instance answering employee questions on connected Slack channels, remain in the free trial period for now, with no end date announced. That means teams whose agents mainly field requests on Slack will pay nothing for the time being, while those making heavy use of agents inside ChatGPT itself will feel the change starting on the first day of July.

For finance and IT departments, this means running a quick pilot before signing long-term corporate agreements. Companies with engineering teams heavily relying on Codex-based tasks should test actual credit consumption before committing to a specific budget, since variable billing is hard to forecast without pilot data.

Implications for businesses in Poland

For Polish companies using ChatGPT Business or Enterprise, the change means that automating internal processes, previously available essentially for free within the subscription, now generates an additional, hard-to-predict variable cost. Companies that in recent months got employees used to relying on agents for accounting work or handling support tickets will have to either scale back usage or accept higher bills without full transparency on the final rate.

OpenAI's decision fits a broader market trend in which model providers are shifting from flat subscriptions to billing based on actual compute usage, shifting part of the cost risk of generative AI from labs onto corporate customers. As memory and compute prices keep rising amid the AI boom, similar changes to billing models could spread to other providers in the coming months.

Sources: Tech Times (techtimes.com), OpenAI Help Center (help.openai.com), 9to5Mac (9to5mac.com)