OpenAI Adds Parallel Agents to ChatGPT Pro, Letting GPT-6 Handle Tasks Simultaneously

OpenAI updated GPT-6 Astra Pro inside ChatGPT on October 16, 2026, enabling multiple agents to work on separate subtasks of a single complex request simultaneously. Eligible Pro subscribers can now follow each agent's progress directly in the conversation window, then open any completed subtask to inspect the output independently. The update also delivers faster responses overall and tighter integration with ChatGPT's plugin and tool ecosystem, which now encompasses more than 4,000 third-party apps. OpenAI confirmed the update via its official X account on October 16.

How Parallel Agents Work Inside the Chat Tab

Before October 16, GPT-6 Astra Pro handled complex requests sequentially: one agent, one chain of steps, one output. The update introduces a new execution model within the Chat tab. When GPT-6 Astra Pro receives a multi-part request — competitive analysis split by market segment, or parallel code-generation and documentation work — it can decompose the goal into concurrent workstreams and assign one agent per lane. Agents run in parallel rather than queuing behind each other, and users see each workstream's progress as it unfolds in the same conversation window.

When an agent completes its portion, the user can open that subtask and review the output independently, before a parent orchestrator assembles the results into the final response. OpenAI's release notes say the update also improves plugin and tool use, suggesting each parallel agent can invoke tools independently rather than routing through a single sequential call chain.

One scope boundary: the October 16 API changelog records only voice-related updates, not a new parallel-agent endpoint. The parallel-agent capability is a Chat product feature, not an exposed API primitive. Developers building orchestration workflows on the OpenAI API cannot directly invoke the same coordination layer; that distinction separates this from an infrastructure change and places it firmly in the consumer-product tier.

GPT-6 Astra Pro's Orchestration Model and Its Design Trade-Offs

GPT-6 Astra is the model underlying ChatGPT's Pro tier since its September 3, 2026 release, also powering Codex and the Dots agents introduced at DevDay later that month. In the Chat tab, the model appears to perform the decomposition decision itself — identifying which subtasks are sufficiently independent to run in parallel — rather than requiring the user to specify an agent graph.

This differs from API-based multi-agent orchestration, where a developer writes the decomposition logic, agent topology, tool allocation, and retry behavior in application code. The Chat product trades away that configurability — the number of agents, their tool assignments, fallback logic — in exchange for making parallelized work accessible to non-technical users without orchestration code. A user interacting through the ChatGPT interface cannot currently instruct the model to spawn a specific number of agents, allocate particular tools to a particular agent, or define failure-recovery paths.

The trade-off is deliberate. Non-technical Pro subscribers can now route a multi-part research or analysis task through simultaneous agents from a chat prompt, a capability that previously required either OpenAI's Assistants API, a third-party orchestration framework such as LangGraph or AutoGen, or a specialist agentic product.

Three Agentic Consumer Releases in Six Weeks

October 16 is the third significant agentic consumer release from OpenAI in six weeks.

On September 29 at DevDay, OpenAI launched Dots — persistent, always-on agents running GPT-6 Astra on dedicated cloud computers, capable of continuing work between conversations and reaching more than 4,000 apps through the plugin ecosystem. Dots are available to Pro and Business Premium subscribers. Dots represent a long-horizon asynchronous execution model: a user delegates a multi-session background task and the agent returns to it between conversations.

The parallel-agent feature in Chat occupies a different execution niche. These agents are session-bound: they fan out from a single prompt, run concurrently, and return results within the same conversation. They do not persist between sessions. The two capabilities are complementary — Dots for asynchronous background work across sessions, parallel Chat agents for synchronized fan-out within an active session.

OpenAI also introduced a GPT-5.6 "ultra" setting in July 2026 that ran up to four agents in parallel for coding tasks within Codex, raising GPT-5.6 Sol's Terminal-Bench 2.1 score from 88.8% to 91.9% in that context — company-reported figures that have not been independently reproduced. The October 16 release extends the parallel execution model from a specialized coding workflow into the general ChatGPT Chat interface.

Anthropic Has Parallel Agents but Not in a Consumer Chat Interface

The most direct competitive precedent is Anthropic's Claude Code Agent Teams, launched February 5, 2026, alongside Claude Opus 4.6. In that system, a lead agent in the Claude Code CLI can delegate subtasks to multiple teammate agents running concurrently, enabling simultaneous research, code generation, and debugging within one session. Anthropic documented an internal use case in which sixteen Claude Opus 4.6 agents worked in parallel to build a 100,000-line Rust-based C compiler across nearly 2,000 sessions, at an estimated API cost of approximately $20,000 — figures Anthropic has stated but that have not been independently reproduced.

The deployment surface differs significantly from OpenAI's October 16 update. Claude Code Agent Teams require terminal-based installation, configuration, and explicit orchestration setup — a developer tool aimed at engineers. OpenAI's parallel Chat agents are available through the ChatGPT browser or app interface without installation or code. For non-technical Pro subscribers, that surface difference is the key point: parallelized AI work is now accessible from a chat prompt rather than from a terminal window.

Anthropic advanced its parallel-agent capabilities on the developer side with Claude Code Projects, launched September 17, 2026, in limited beta for Pro and Max users, with access extended to all waitlisted Pro and Max users by October 9. That release narrows the gap on the developer tool side. Neither Claude Code Projects nor the original Agent Teams brings parallel orchestration to Anthropic's consumer chat interface, which remains sequential.

Read more: Claude Code's parallel agent architecture for developers

Usage Metering and Rollout Scope Remain Unresolved

OpenAI has not disclosed how parallel agent runs interact with Pro subscription usage limits. Whether each sub-agent draws from the monthly usage cap independently, whether the agents share a combined pool, or whether the feature has separate metering has not been addressed in the release notes or any follow-up communication as of October 19. This is a material open question for Pro users planning to use parallel execution on intensive tasks: if each of three simultaneous agents counts as a full usage draw, the effective per-task cost of the feature could be three times a standard single-agent request.

The rollout is also staged rather than universal. The October 16 release reaches eligible Pro users in the Chat tab; Team, Business, and Enterprise accounts are not included in this wave, and OpenAI has not announced a timeline for those plans. Not all Pro accounts received the feature simultaneously — the rollout is incremental.

Early user reactions since October 16 have been broadly positive on the speed improvement and the ability to watch multiple workstreams simultaneously, with a subset of users reporting that the feature is more conservative than expected about which tasks it accepts for parallelization, and that per-agent decision transparency falls short of what they expected.

The October 16 update advances OpenAI's stated direction of making multi-agent AI coordination a standard interface choice rather than an infrastructure investment. Whether that shift reaches organizations at scale depends on two unresolved decisions: how OpenAI meters parallel runs against Pro limits — which will determine whether intensive parallel use is financially practical — and when the feature extends beyond individual Pro accounts to Team and Enterprise, where multi-person workflows have the clearest need for parallelized AI workstreams. Those two decisions will determine whether October 16 marks a broad behavioral change in how knowledge workers use ChatGPT, or the opening stage of a more measured rollout.