OpenAI Launches Persistent Agents, Computer Use, and Cheaper Models at DevDay 2026

OpenAI used its annual developer conference to introduce a lineup of capabilities emphasizing persistent, multi-task automation. The announcements signal a shift away from discrete model interactions toward agents capable of sustained operation across software and web interfaces.

Agent Infrastructure and Computer Use#

The centerpiece for developers is expanded Agents API functionality that now includes computer use—allowing applications to interact with software through graphical interfaces rather than just API endpoints. The platform now handles multi-agent coordination, tool discovery, and execution infrastructure management directly, with the capability accessible via API as well as in Codex and ChatGPT Work.

OpenAI also introduced Dots, persistent agents designed to tackle ongoing work. These agents operate alongside ChatGPT Space, a collaborative workspace where teams and autonomous systems share context and coordinate tasks.

New Models and Cost Efficiency#

GPT-6.1 Sol targets coding, software operation, and professional workflows. According to OpenAI, the model reaches performance comparable to GPT-6 Astra on several benchmarks while costing only one-fifth as much. Input tokens are priced at $0.10 per million when using caching, making it a significant cost reduction for cost-conscious builders.

The Decisions API, now in limited preview, handles classification and routing tasks using a smaller Luna-class model. It picks from predefined options based on text or image context, automating common branching logic in workflows.

Developer Tools and Extensibility#

Cloud-based Codex environments complement local setups, enabling remote task execution across devices. The Codex command-line interface gained voice input and an /agents interface for spawning and monitoring multiple parallel tasks. A new code-review feature integrates with GitHub and GitLab, analyzing pull requests and merge requests for defects, while Codex Security Cloud scans repositories for vulnerabilities and proposes fixes.

ChatGPT's plugin ecosystem expanded with sidebar experiences, interactive panels, and custom file viewers. Plugins now respond to external events through the MCP Events specification, enabling automations that trigger based on activity in connected applications.