ChatGPT parallel agents are now rolling out to Pro users. GPT-6 Pro can split one request into several tasks and run them at the same time. OpenAI announced it on X on October 16, 2026, and early testing says the speed gain is real but small.
Here is the short version:
- Announced: October 16, 2026, for Pro users in the Chat tab
- Several agents handle different parts of one task, side by side
- You can watch each agent and open finished tasks to check the work
- OpenAI also says responses are faster and plugin and tool use is better, with no numbers given
That list covers the headline. The details below decide whether it is worth your time.
How ChatGPT parallel agents actually work
Think of it like handing a research brief to a small team instead of one person. GPT-6 Pro breaks your request into pieces, and each agent takes a piece.
While they work, you can follow every agent’s progress inside the conversation. When one finishes, you open its task and read what it did. That makes it easier to spot a dropped caveat or a weak source before you trust the final answer.
One catch: OpenAI hasn’t explained how Chat does this internally. Independent testers describe it as a conceptual model, not a confirmed design, according to Kingy’s hands-on test.
Is it actually faster? What early testing shows
Kingy ran a three-part research brief twice in each mode on a single eligible account. The parallel runs finished a bit sooner on average.
| Mode | Average time | What happened |
|---|---|---|
| One agent (sequential) | 4 min 51 sec | Baseline run |
| Three agents (parallel) | 4 min 25 sec | About 9% faster, 26 seconds saved |
| Run-to-run spread | -10 sec to +62 sec | One parallel run was slower than its control, the other 62 seconds faster |
Treat that as a hint, not proof. It was four runs on one account, and accuracy looked about the same either way. Speed gains will likely grow with bigger jobs, but nobody has shown that yet.
Who can use GPT-6 Pro right now?
OpenAI says the rollout is going to eligible Pro users, so you may not see it today even on the right plan. Based on the GPT-6 plan guide as summarized by Kingy, access looks like this:
| Plan | GPT-6 Pro in Chat |
|---|---|
| Pro ($100) | Listed |
| Pro ($200) | Listed |
| Business | Listed, depends on workspace settings |
| Enterprise | Listed, depends on workspace settings |
| Plus | Not listed for the regular Chat tab |
If you are on Plus, one commenter has already asked for the feature, but OpenAI has shared no plans to expand it. For background on how OpenAI is reshaping the app around GPT-6, see our look at the GPT-6 intelligent UI in ChatGPT.
Which jobs suit parallel agents (and which don’t)
Parallel work shines when the pieces don’t depend on each other. Good fits include:
- Researching three vendors or products at once
- Comparing several documents against one checklist
- Auditing a list of pages, sources or claims
These jobs split cleanly, so each agent can finish without waiting on another. That is where you save time.
Short questions, step-by-step tasks and anything where one result feeds the next are better left to a single agent. Extra agents add overhead without helping. If you build slide decks from the results, our guide to ChatGPT Slides access and export shows what to do next.
How to get more out of ChatGPT parallel agents
Write your request so the pieces are obvious. Name each lane, such as pricing, reviews and risks, and say what a finished answer looks like for each one.
Then open the finished tasks before you read the summary. Check that each agent kept the caveats and sources you asked for. A merged answer can smooth over a weak finding, and the task records are where you can catch that.
Keep an eye on your usage too. Because OpenAI hasn’t published how parallel runs are counted, run a couple of small jobs first and see how your allowance behaves before you throw a huge brief at it.
What OpenAI still hasn’t told us
The announcement leaves several practical questions open. OpenAI hasn’t said how many agents can run at once, or when ChatGPT decides to split a task on its own.
It also hasn’t said how parallel runs count against your Pro allowance. Testers saw no per-run usage figure, and the activity panel still said “No outputs yet” after the agents finished. Plugins stay under your workspace’s approval rules, so extra agents don’t gain extra permissions.
Even the name is muddled. The X post says “GPT-6 Astra Pro,” while the Help Center says “GPT-6 Pro, powered by GPT-6 Astra.”
Frequently Asked Questions
Are ChatGPT parallel agents free?
No. They are part of GPT-6 Pro, which is a paid Pro feature. OpenAI has not announced a free or Plus version.
How many agents can run at once?
OpenAI hasn’t said. Kingy’s test used three agents, but that is not a confirmed limit.
Do parallel agents make answers more accurate?
Not so far. The one early test found no clear accuracy gain over a single agent.
Can I check what each agent did?
Yes. You can follow progress live and open each completed task to review its work, including sources and results.
Should you care yet?
If you already pay for Pro and do multi-part research, try it on your next big brief and compare the time yourself. If you don’t, there is no rush. A 9% gain from four test runs is a nice bonus, not a reason to upgrade, and the biggest benefit for now is being able to inspect each agent’s work.
Leave a Reply