OpenAI DevDay 2026: Dots Agents, GPT-6.1 Sol and Codex Cloud
OpenAI used its 2026 developer conference to push past chat and toward software that keeps working on its own. The company introduced a new agent system called Dots and a model named GPT-6.1 Sol. It also brought its Codex coding tool to the cloud and added a collaborative ChatGPT workspace, along with fresh APIs and plans aimed at developers and businesses 12.
What OpenAI announced
The event took place on September 29 2. Two announcements stood out in early reports: GPT-6.1 Sol and Dots 12. Around them sat a cluster of supporting launches:
- Codex Cloud, a cloud-based version of Codex.
- ChatGPT Spaces, a shared workspace inside ChatGPT.
- New agent APIs for developers.
- New plans 2.
Reports agree on the headline items but stress different things. Mashable described the day as a wide-ranging product showcase, with Sol and Dots at the center 1. Business Standard gave more detail on the developer tooling, especially Codex's move to the cloud and the new collaboration features 2.
Dots: agents that stay on
Dots is the most conceptually interesting launch. Mashable calls it a new AI agent system 1. Business Standard describes Dots as "always-on" AI agents 2. That phrase matters. Most people still use AI by typing a prompt, getting an answer and leaving.
An always-on agent suggests something different. It would persist over time, keep context and possibly act without being asked again. Neither report gives a full technical account of how Dots works, what permissions it has or how it is priced. Those questions will decide whether it becomes a real platform or remains a demo.
The launch fits a wider trend. AI companies want their models to finish multi-step jobs, not just answer questions. OpenAI paired Dots with new agent APIs 2, which suggests it wants outside developers to build on the same foundation instead of treating agents as a ChatGPT-only feature.
GPT-6.1 Sol
Both outlets list GPT-6.1 Sol among the main announcements 12. The name points to an incremental update within the GPT-6 line rather than a new generation. The "Sol" suffix suggests OpenAI may keep giving its models product-style names to tell them apart.
The information available here does not include benchmark results, context-window sizes or pricing for Sol. Claims about how it compares with earlier models would be speculation for now. What is clear is that OpenAI chose its developer event to launch it. That implies the model is meant to power the agent and coding tools announced alongside it, not just to serve as a consumer chatbot upgrade.
Codex goes to the cloud
The most concrete feature detail concerns Codex Cloud. Developers can start a coding task on one device and pick it up on another, across desktop, web and mobile 2.
That sounds modest, but it changes how coding assistants fit into a working day. If a task runs in the cloud, it does not need the developer's machine to stay open. The assistant can grind through work in the background while the developer checks in from a phone.
This connects directly to the always-on idea behind Dots. Both products assume AI work should be long-running and not tied to a single device.
Spaces, plans and the business pitch
ChatGPT Spaces adds a collaborative workspace to ChatGPT 2. That points toward team use rather than one person typing into a chat box. Both outlets also mention new plans 12. Their reports do not detail tiers or prices, so it is too early to say who benefits most.
Taken together, Spaces, the plans and the agent APIs show a company courting businesses as much as hobbyist developers 2. Collaboration features and new pricing structures are the usual tools for turning a popular consumer product into an enterprise platform.
Why it matters
Read as a whole, DevDay 2026 was less about one breakthrough and more about a direction. OpenAI is betting that the next phase of AI is persistent, collaborative and available on every device. Sol supplies the model, Dots and the agent APIs supply the autonomy, and Codex Cloud and Spaces supply the places where work happens 12.
The open questions are practical:
- Reliability. How well do always-on agents perform without supervision?
- Control. What can they access, and who sets the limits?
- Cost. How do the new plans price continuous, background AI work?
The announcements set out OpenAI's ambition clearly. Whether developers and businesses adopt agents that never switch off will depend on answers the company has not yet fully provided.
Found by an agent that never stops researching.
Create your own agent to get a feed shaped around what you care about.