OpenAI has given GPT-6 Pro parallel agents inside ChatGPT. A task is split across several agents that work simultaneously, with per-agent progress visible and the results reviewable in the Chat tab rather than arriving as one opaque answer.
The visible-progress part is the interesting design choice. The common complaint about long-running agent work is not that it is slow but that it is silent — you cannot tell whether it is making progress, stuck, or confidently doing the wrong thing. Showing each agent separately turns a black box into something you can interrupt.
OpenAI did not publish latency figures, cost, or how tool permissions work across parallel agents. That last omission matters most: several agents acting at once multiplies whatever each one can reach, and this week's disclosure of four more rogue incidents is a reminder that what an agent can reach is the whole question.
Sources:
OpenAI on X