ChatGPT Dots Just Got a Lot Smarter: How OpenAI’s Always-On Agents Are Learning to Pick Their Own Brains

 

By Diablo Tech Blog | November 4 2026 

OpenAI’s ChatGPT Dots launched in late September 2026 as one of the most ambitious consumer AI features in years. These persistent, always-on personal agents come with their own cloud computers and browsers, connect to thousands of apps, remember context across conversations, and keep working on your goals even when you’re offline. They were designed to move beyond the “prompt and wait” paradigm into something closer to a proactive digital teammate.

This is not a minor polish. It points to a deliberate shift in how OpenAI thinks about agent intelligence: less user micromanagement of models and more autonomous orchestration by the agent itself.

A Quick Refresher: What Dots Actually Are

Dots are not simply ChatGPT with memory. Each Dot is a persistent agent powered (at launch) primarily by GPT-6 Astra. It runs on its own isolated cloud computer with a browser, can connect to more than 4,000 apps through OpenAI’s plugin ecosystem, and continues working toward goals between conversations. Users can interact with their Dot through ChatGPT (desktop, web, and increasingly mobile), Slack, Microsoft Teams, and—as of late October 2026—via text on iMessage or RCS in a limited U.S. Pro beta.

Key characteristics that set them apart from earlier agent experiments:

• Always-on persistence: Tasks and research continue in the background.
• Dedicated compute: Each Dot has its own environment rather than piggybacking on the user’s device.
• Memory and learning: They retain context about your preferences, projects, and past interactions and improve from feedback.
• Permission controls: Users set rules for what the Dot can do autonomously, what requires approval, and what is off-limits. Sensitive actions (such as changing passwords) remain with the human.
• Availability: Rolling out to ChatGPT Pro (outside the EEA, UK, and Switzerland at launch) and Business Premium users. Enterprise, Edu, and Healthcare get an admin-enabled beta. The first Dot is included at no extra cost on eligible plans.

In short, Dots aim to be the difference between an AI that answers questions and one that owns ongoing responsibilities.

Understanding the Models: Astra vs. Terra

To appreciate why this multi-model approach matters, it helps to understand the current GPT-6 family positioning (as of early November 2026):

• GPT-6 Astra: OpenAI’s top-tier frontier model at launch for Dots. It is positioned as the most capable and best-aligned option for demanding professional work, complex reasoning, coding, computer use, and high-stakes tasks. It is also the most expensive.
• GPT-6.1 Terra: Released on October 26, 2026, Terra is a fast, efficient mid-tier model optimized for visual reasoning, precise computer use, and natural writing. OpenAI reports that it matches or approaches Astra-level performance on several agent and computer-use benchmarks at roughly one-tenth the per-token cost of Astra in some comparisons, while completing many tasks more than three times faster. It also carries a lower price point that helps usage allowances stretch further.

Terra is not a pure replacement for Astra. OpenAI still recommends Astra (or higher-tier options like Sol in certain contexts) for the hardest end-to-end reasoning and coding jobs. But for high-volume everyday agent work—research, form filling, screen interaction, drafting, monitoring—Terra offers a compelling balance of speed, capability, and cost.

By giving Dots access to both (and the ability to mix them), OpenAI can optimize for quality where it matters most and efficiency where volume or latency is the priority. A single complex project might use Terra for rapid iterative research and screen navigation, then switch to Astra for final synthesis or high-stakes decision support.

Why Automatic Model Selection Matters

Most AI users still treat models as static tools they must choose. Power users develop heuristics (“use the expensive one for hard reasoning, the fast one for drafting”). Casual users often stick with the default and accept whatever quality-cost tradeoff comes with it.

Dots flip this. The agent becomes the router. In principle, this should:

• Reduce cognitive load on the user.
• Improve overall task success by matching capability to need.
• Stretch plan allowances further by avoiding unnecessary use of premium models.
• Enable more sophisticated multi-step workflows that require different strengths at different stages.

There are open questions, of course. Will users be able to see which model was used for which step? Will there be an override? Early reactions online already include requests for transparency and manual control. OpenAI has not publicly detailed the exact implementation or user-facing controls at the time of the Android Authority report.

The broader trend is clear: OpenAI is moving the intelligence about how to use the models into the agent layer itself. This is consistent with the company’s longer-term direction of reducing configuration surface area while increasing agent autonomy.

Practical Implications for Users

For people already using Dots (or waiting for rollout), the update should feel like a quiet upgrade in capability and responsiveness rather than a new feature that requires learning. Tasks that previously felt constrained by a single model’s strengths or by cost/latency tradeoffs may become smoother.

Specific areas likely to benefit:

• Long-horizon projects (vacation planning, research synthesis, multi-app workflows) that mix research, analysis, and action.
• High-volume monitoring or routine work where Terra’s speed and efficiency shine.
• Mixed-modality tasks involving screenshots, documents, or visual reasoning.
• Scaling access: Cheaper, faster models make it more feasible for OpenAI to expand Dots beyond the initial Pro and Business Premium cohorts.

Safety and control remain important. Dots already include auto-review for risky actions and user-configurable permission levels. Expanding the model pool does not remove the need for those guardrails; if anything, it increases the importance of clear observability into what the agent is doing and why.

The Bigger Picture: Agents That Orchestrate Themselves

The November update is one data point in a larger shift. OpenAI, Anthropic, Google, and others are all racing to move from single-model chat interfaces toward persistent agents that manage tools, memory, and compute on the user’s behalf. Dots were already a strong statement of that direction. Giving them the ability to choose and switch models is a logical next refinement.

It also reflects economic reality. Frontier models remain expensive to run at scale. Intelligent routing between capability tiers is one of the more practical ways to deliver high-quality agent experiences without making every interaction cost as much as the most powerful model.

Whether this particular implementation lands well will depend on execution details that are still emerging: transparency, reliability of the routing decisions, latency of model switches, and how well the system handles edge cases. Early users will be the ones to stress-test it.

Looking Ahead

OpenAI has signaled that more models will join the Dots lineup over time. Combined with ongoing improvements in mobile setup, texting, widgets, plugin ecosystem, and specialist/team Dots, the platform is clearly intended to evolve rapidly.

For now, the headline is straightforward and significant: ChatGPT Dots are no longer limited to a single brain, and they are starting to decide for themselves which brain to use. That is a meaningful step toward the kind of proactive, low-friction AI assistance OpenAI has been describing since the product launched.

If you have access to Dots, the practical advice is simple—keep using them as you have been. The system is designed so you should not need to change your behavior. The intelligence about model choice is moving behind the scenes, where many users prefer it to stay. The real test will be whether the results feel consistently smarter, faster, and more reliable without requiring extra effort from the human on the other side of the conversation.

To read more such articles, hit the dotted line on the right and follow us !! Never miss what's new, what's happening, what's released and our in-depth analysis on the latest and greatest tech has to offer