OpenAI DevDay 2026: Dots Agents, GPT-6.1 Sol and New Plans

OpenAI used its DevDay 2026 event to push hard on one idea: AI that keeps working after you stop typing. The company introduced a stack of products aimed at developers and businesses, led by a new agent system called Dots and a new model, GPT-6.1 Sol 12. The event took place on September 29 2. Alongside those headliners came cloud-based Codex, new agent APIs, a collaborative ChatGPT workspace called Spaces, and new plans 2.

Coverage agrees on the two centerpiece announcements. Mashable describes the event as a broad product dump, with Sol and Dots as the anchors 1. Business Standard gives a fuller inventory of developer tooling and more detail on how the pieces work 2.

Dots: agents that don't clock out

The most consequential announcement is Dots, which OpenAI pitches as always-on AI agents. The core promise is persistence. A Dot keeps pursuing a task after the initial instruction rather than stopping after a single response 2.

The architecture shows how OpenAI is thinking about this. Each Dot runs on GPT-6 Astra and gets its own cloud computer and browser 2. That lets it work through connected tools and services as it moves toward a goal 2. In practice, Dots are less a chatbot feature than a hosted worker. Each has a sandboxed machine, web access and integrations, and runs on OpenAI's infrastructure instead of the user's device.

That framing matters. Many agent products so far have been limited by session length, by needing a human to keep a browser tab open, or by fragile local setups. Giving each agent a persistent cloud environment is OpenAI's attempt to remove those constraints. It also raises the questions that always follow autonomous software: how users supervise long-running tasks, what permissions agents hold over connected accounts, and how errors get caught when no one is watching in real time. The available reporting does not say how OpenAI addresses those issues. They will likely decide whether businesses trust Dots with real workloads.

GPT-6.1 Sol: near-flagship performance, lower cost

The second headline item is GPT-6.1 Sol, the newest member of the GPT-6 family 12. OpenAI says Sol performs close to GPT-6 Astra at a lower cost. It is aimed at coding, computer use and professional workloads 2. Business Standard also mentions an "Ultrafast" offering alongside Sol, though the reporting gives little detail on it 2.

This follows a familiar pattern in AI releases: a flagship model sets the capability ceiling, then a cheaper sibling captures most of that capability for everyday use. Sol's focus areas line up neatly with Dots. Agents that operate computers and write code need a model that is good at computer use and coding, and cheap enough to run for long stretches. If Astra is the premium engine behind Dots, Sol looks positioned as the cost-efficient workhorse for developers building their own agentic products. That reading is an inference, but it fits the emphasis OpenAI placed on both launches.

The supporting cast: Codex Cloud, agent APIs and Spaces

Beyond the headliners, OpenAI rolled out several tools that round out its developer pitch. Codex, its coding product, now runs in the cloud 2. New agent APIs let developers build their own autonomous systems 2. ChatGPT Spaces is a collaborative workspace inside ChatGPT, which suggests a push toward team use rather than solo chats 2. Both outlets also point to new plans 12, though the reporting gives no specifics on pricing or tiers.

Taken together, these announcements show a consistent strategy. Codex Cloud moves coding agents off local machines, just as Dots does for general tasks. The agent APIs open the same capabilities to third parties. Spaces gives teams a shared place to work with these systems. New plans likely serve as the commercial wrapper for all of it.

What it adds up to

The clearest takeaway from DevDay 2026 is that OpenAI now treats agents as its main platform, not an experimental add-on. The company is building at several layers at once: a persistent agent product (Dots), a premium model to power it (Astra), a cheaper model for scale (Sol), cloud infrastructure for coding (Codex Cloud), and APIs so developers can build their own versions.

The sources mostly agree and differ chiefly in depth. Mashable offers the high-level picture, while Business Standard supplies the technical specifics. What neither yet provides is evidence of how these systems perform under real business conditions. Until then, DevDay 2026 is best read as a statement of direction: OpenAI wants its AI to run continuously in the cloud, doing work on users' behalf.