OpenAI DevDay 2026: Computer Use, Cloud Codex and GPT-6.1 Sol

OpenAI used DevDay 2026 to push a clear message to developers: the company wants to own more of the stack that agents run on, not just the models that power them. The announcements span a new model, an expanded Agents API with computer use, cloud-hosted Codex environments, a new Decisions API, and a broader plugin system for ChatGPT. 12

What OpenAI announced

The headline list includes GPT-6.1 Sol, computer use for the Agents API, cloud-based Codex environments, a Decisions API, and new plugin capabilities inside ChatGPT. 12 InfoQ's coverage gives these items equal billing at the top, but the deeper detail concentrates on three areas: agents, Codex, and plugins. 2 Public specifics about GPT-6.1 Sol and the Decisions API are comparatively thin, so their capabilities and positioning are harder to assess. Developers should treat them as names to watch rather than features to plan around.

Agents that can operate software

The most consequential change for agent builders is computer use in the Agents API. Applications can now drive software through its graphical interface rather than relying solely on structured APIs. 2 That matters because a huge share of enterprise software, from legacy internal tools to SaaS dashboards, has no clean programmatic interface. An agent that can click through a UI can, in principle, automate work that previously required custom integration or a human operator.

The Agents API also absorbs several capabilities that originated in Codex: multi-agent coordination, tool search, tool calling, and context compaction. 2 Context compaction is the quieter but important piece, since long-running agents tend to exhaust their context windows, and compaction is how they keep working without losing the thread. OpenAI says it manages the underlying execution infrastructure, and the functionality is available through the API as well as in Codex and ChatGPT Work for selected plans. 2

The infrastructure point is the strategic tell. By hosting execution, OpenAI removes a real burden from developers, who no longer need to stand up sandboxes or orchestration layers themselves. It also deepens dependence on OpenAI's platform, a trade-off teams will weigh differently depending on their compliance and portability needs.

Codex moves to the cloud

Codex can now run in cloud environments as well as on local machines, which lets developers start remote tasks from other devices. 2 The Codex CLI gains voice input and an /agents interface for delegating and monitoring several tasks at once. 2 Taken together, these features reframe Codex less as a coding assistant at your elbow and more as a pool of workers you dispatch and supervise.

Two additions target the software lifecycle beyond writing code. A new code-review workflow analyzes diffs and flags potential issues in GitHub pull requests and GitLab merge requests. 2 Codex Security Cloud scans repositories and incoming commits, investigates findings, de-duplicates them, and prepares fixes. 2 Supporting both GitHub and GitLab is a pragmatic choice that widens the addressable audience. The security tooling's emphasis on de-duplication and triage, not just detection, speaks to a familiar complaint about scanners that bury teams in noisy alerts.

ChatGPT plugins become application surfaces

On the ChatGPT side, developers can now build sidebar experiences, interactive conversation panels, and custom file viewers. 2 Plugins can also react to events through the proposed MCP Events specification, so automations can fire when something happens in a connected application. 2

This is a shift from plugins as request-response tools toward plugins as persistent UI and event-driven workflows. The reliance on a proposed MCP extension is notable. It ties OpenAI's plugin roadmap to the Model Context Protocol ecosystem, but because the specification is still a proposal, developers building on it should expect some churn before it stabilizes.

Why it matters

The throughline across DevDay 2026 is consolidation. Features that once lived in separate products, such as Codex's multi-agent tooling, are being folded into the general Agents API. Execution is moving onto OpenAI-managed infrastructure, and ChatGPT is becoming a host for richer third-party interfaces. 2 For developers, the upside is less plumbing and faster paths from prototype to working agent.

The caveats are real, though. Several capabilities are gated to selected plans, computer use introduces new reliability and security questions whenever an agent acts through a UI, and key pieces like the Decisions API remain lightly documented in public. 2 The practical takeaway is to experiment early with computer use and cloud Codex, where the details are concrete, and hold firmer commitments until the newer model and APIs are better understood.