GPT-6 Pro parallel agents are now live in ChatGPT: OpenAI has upgraded its top model so that a single request can be split across several AI agents working at the same time. The company announced the change on October 16, saying the update also brings faster responses and better use of plugins and tools.
An AI agent is a model that carries out tasks on its own, such as searching, reading files or running code, rather than only replying with text. Running several agents in parallel means different parts of a job can be done at once instead of one after another.
How the agents work
Instead of one model working through a task step by step, GPT-6 Pro can hand different parts of the job to multiple agents at once. Users can follow each agent’s progress inside the conversation and open finished tasks to see what that agent actually did. The feature is live in the Chat tab for eligible users.
OpenAI has not published figures on how much faster the new setup is, how many agents can run at once in a single chat, or how multi-agent work counts against usage limits. Those details will matter to heavy users, because several agents working at once may use up a plan’s allowance more quickly.
Who gets it
Access is limited to the higher tiers. GPT-6 Pro is available on the $100 and $200 Pro plans, as well as Business and Enterprise, and is not included in the cheaper Plus plan. The rollout is still reaching eligible accounts.
The naming is a little muddled. OpenAI’s help centre describes GPT-6 Pro as powered by GPT-6 Astra, the most capable model OpenAI has released, which first launched in September, as we reported in our story on the GPT-6 Astra launch.
Part of a wider agent push
The new feature carries the multi-agent idea into the general chat window. Earlier this month OpenAI launched GPT-6 with Intelligent UI, which lets ChatGPT build interactive tools such as charts and calculators instead of replying with plain text. Together, the two changes show OpenAI moving ChatGPT from a chatbot towards a workspace where AI does larger pieces of work.
Rivals are moving in the same direction. Google this month introduced a Gemini agent that can create sub-agents and keep working in the cloud for hours or days, and Anthropic has pushed its Claude models as sub-agents for coding work. Splitting a big task among several cooperating agents is becoming a standard design across the industry.
Where parallel agents help
Splitting work among agents should help most with tasks that divide cleanly, such as researching several topics at once, comparing multiple documents or checking many sources in parallel. Sequential jobs, where each step depends on the last, are unlikely to gain as much, because later agents have to wait for earlier results anyway.
Coordination is the hard part. When several agents work on one request, their results have to be combined into a single answer, and mistakes by one agent can carry into the final result. That is why the ability to open each agent’s finished work matters: it lets users check what was actually done rather than trusting only the summary.
The oversight question
Visibility may matter more than speed. Adding agents does not automatically give them extra permissions, but more agents still means more activity to review. AI companies have faced growing scrutiny this year over agents that took actions their developers did not intend, so tools that show what each agent did are a useful safeguard. How much oversight they provide in practice, when a single request launches several agents at once, remains to be seen.
Why it matters for businesses
For business users on Pro, Business or Enterprise plans, parallel agents could shorten research and analysis work, such as market scans, document reviews or competitor comparisons. The practical advice is to start with tasks that split naturally into independent parts, check the work of each agent before relying on the result, and watch usage limits, since parallel work may consume them faster.
Our take
Parallel agents are a logical next step for ChatGPT and a clear sign that the AI race is shifting from smarter answers to getting more work done. The feature is only useful if people can trust the combined result, and OpenAI’s decision to let users inspect each agent’s work is the right one. But with no published data on speed, limits or accuracy, the benefits are still a promise. Businesses should treat it as a capable assistant team that still needs a human manager.