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

What OpenAI Announced

OpenAI used DevDay 2026 to roll out a broad set of developer-facing updates. The headline items were a new model, GPT-6.1 Sol; computer use in the Agents API; cloud-hosted Codex environments; a new Decisions API; and expanded plugin capabilities for ChatGPT 12. Both accounts agree on that list of announcements. Most of the available detail concerns the agent tooling, the Sol model update, and Codex. The Decisions API and the ChatGPT plugin changes were named without much further description, so their practical shape is still unclear.

The overall direction is easy to read. OpenAI is betting heavily on agents, meaning software that takes actions rather than just answering prompts. It also wants to own more of the infrastructure those agents run on.

Agents That Can Operate Software

The most consequential change for application builders is likely the addition of computer use to the Agents API. This lets applications drive software through graphical interfaces rather than relying only on structured APIs 2. Agents can now, in principle, work with tools that were never designed to be automated, including legacy desktop applications, web dashboards, and internal portals.

The Agents API also absorbs several capabilities that originated in Codex. These include multi-agent coordination, tool search, tool calling, and context compaction 2. OpenAI says it will manage the execution infrastructure underneath 2. The features are available through the API and in Codex and ChatGPT Work for selected plans 2.

This managed approach involves a clear tradeoff. Developers no longer have to orchestrate sandboxes, long-running sessions, or memory management themselves, which lowers the barrier to shipping agentic products. The cost is that more of the stack sits inside OpenAI's platform, which deepens dependence on a single vendor. That is a reasonable price for teams that want to move quickly. Organizations with strict control or portability requirements may find it harder to accept.

GPT-6.1 Sol: Competing on Price

GPT-6.1 Sol updates GPT-6 Sol and targets coding, computer use, and professional work 2. OpenAI's main claim concerns economics. It says the model comes close to GPT-6 Astra on several evaluations while costing one-fifth of Astra's standard input and output token prices 2. Cached input is priced at $0.10 per million tokens 2. The model is available through the API, ChatGPT Work, and Codex 2.

The positioning is significant. Agentic workloads use a lot of tokens: they loop, re-read context, and call tools repeatedly. For those workloads, per-token cost often matters more than marginal gains on benchmarks. A model that is "nearly as good" at a fraction of the price could become the default for production agents, with Astra-class models kept for the hardest tasks.

The low cached-input price fits this picture, since agents tend to resend large shared contexts. Still, "approaches on several evaluations" is OpenAI's own framing. Developers should test Sol against their own workloads before treating it as a substitute for the premium tier.

Codex Moves to the Cloud

Codex can now run in cloud environments as well as on local machines. This means developers can start remote tasks from other devices 2. The Codex CLI also gained voice input and an /agents interface for delegating and monitoring multiple tasks at once 2.

OpenAI also expanded Codex into review and security. A new code-review workflow examines diffs and flags potential issues in GitHub pull requests and GitLab merge requests 2. Codex Security Cloud scans repositories and incoming commits, investigates findings, removes duplicate results, and prepares fixes 2.

Taken together, these updates push Codex beyond an in-editor assistant toward something closer to a background engineering service. In that model, a developer dispatches tasks, monitors parallel agents, and reviews their output. Support for both GitHub and GitLab suggests OpenAI wants to fit into existing workflows rather than require teams to switch platforms.

Why It Matters

Several threads run through these announcements. Agent capabilities first developed in Codex are being made available to all API users. A cheaper model is tuned for exactly the long, tool-heavy sessions those agents create. And OpenAI is offering managed infrastructure so developers can avoid building it themselves. Each piece supports the others.

The strategic aim appears to be making OpenAI's platform the easiest place to build and run agents end to end, from the model to execution to code review. Whether that works will depend on factors not yet established. These include how reliable GUI-based computer use proves in practice, whether Sol's benchmark parity holds up in real deployments, and what the Decisions API and new plugin capabilities turn out to do. For now, DevDay 2026 clearly signals that OpenAI sees agents, rather than chat, as its next main developer platform.