OpenAI DevDay 2026: GPT-6.1 Sol, Agents API Computer Use

What OpenAI Announced

OpenAI used DevDay 2026 to ship a dense bundle of developer-facing updates. The headline items are a new model, GPT-6.1 Sol; computer use inside the Agents API; Codex environments that run in the cloud; a new Decisions API; and expanded plugin capabilities for ChatGPT 12. Both accounts of the event list the same five pillars. The more detailed coverage concentrates on the agent tooling, the model and Codex, and offers little on the Decisions API or the ChatGPT plugin changes beyond naming them 2. Those two pieces remain largely undefined for now, and developers should wait for documentation before drawing conclusions about them.

Taken together, the announcements share a theme. OpenAI is trying to own more of the stack that sits between a model and finished work: the execution environment, the orchestration layer and the review loop.

Agents That Can Use a Computer

The most consequential change for agent builders is that the Agents API now supports computer use. Applications can drive software through its graphical interface rather than relying only on structured APIs 2. The API also absorbs several capabilities that were previously associated with Codex [2]:

• multi-agent coordination
• tool search
• tool calling
• context compaction

OpenAI manages the underlying execution infrastructure itself 2.

That last point matters as much as the feature list. Running GUI-driving agents reliably means provisioning sandboxes, managing state and handling long sessions that would otherwise blow past context limits. Hosting that work moves complexity off the developer's plate. The trade-off is that more of the agent's runtime sits inside OpenAI's walls. The capability is available through the API and in Codex and ChatGPT Work for selected plans 2, so access is not universal at launch.

GPT-6.1 Sol: Near-Flagship Results at a Lower Price

GPT-6.1 Sol updates GPT-6 Sol and targets coding, computer use and professional tasks 2. OpenAI's pitch is economic. The company says the model approaches 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 offered through the API, ChatGPT Work and Codex 2.

Two caveats apply:

• The benchmark claim is vendor-reported. "Approaches" on "several evaluations" is a careful phrase that leaves room for gaps on other workloads.
• The price matters most for agents. Agentic workflows such as GUI navigation, multi-step coding and repeated tool calls burn through tokens quickly. A cheaper model that is good enough, paired with low cached-input pricing, could make those long-running jobs far more viable than a premium tier would. That link to the computer-use launch looks deliberate.

Codex Moves to the Cloud

Codex can now run in cloud environments as well as on local machines. Developers can 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.

OpenAI also introduced two review and security features [2]:

• Code review: a workflow that analyzes diffs and flags potential issues in GitHub pull requests and GitLab merge requests.
• Codex Security Cloud: a service that scans repositories and new commits, investigates findings, removes duplicates and prepares fixes.

This pushes Codex from a coding assistant toward a participant in the software delivery pipeline. Deduplicating findings and drafting fixes speaks to a real complaint about automated security scanners, which is that they bury teams in noisy alerts. Whether Codex Security Cloud actually reduces triage burden, rather than adding another stream of suggestions to review, is something teams will have to measure for themselves.

Why It Matters

Viewed as a whole, DevDay 2026 reads less like a model launch and more like a platform consolidation. Codex features migrating into the general Agents API suggest OpenAI wants one agent substrate that serves its own products, including Codex and ChatGPT Work, as well as third-party developers 2. A cheaper capable model makes that substrate affordable to run at scale 2. Cloud execution and managed infrastructure make it easier to adopt 2.

For developers, the practical calculation is convenience versus control. Managed execution, built-in context compaction and hosted Codex environments remove real engineering work. They also deepen dependence on a single vendor's runtime, pricing and plan-tier gating. The strongest near-term case is probably GPT-6.1 Sol for cost-sensitive agent workloads, provided its benchmark claims hold up in independent testing. The Decisions API and ChatGPT plugin updates are the open questions to watch as details emerge 1.