OpenAI’s Codex updates at DevDay focus on where the coding agent runs rather than what it writes. Developers can now keep reusable cloud development environments instead of treating each remote task as a disposable sandbox.
The environments are configurable and shareable, so a team works from one approved set of settings and permissions. Codex is reachable from a laptop, a phone or the cloud. The refreshed command line tool adds voice control and a new agents view for tracking several tasks at once, along with a cleaner terminal interface.
Security scanning is now a first-class feature. Codex Security Cloud watches entire GitHub repositories on demand, on a schedule, or whenever new commits land, then investigates findings, removes duplicates and prepares fixes while the developer’s machine is closed. Users get access to the Daybreak Blue defensive model tier without applying separately.
The API side adds a Decisions API that uses GPT-6 Luna to answer a defined set of questions with predefined responses, useful for routing and classification. The Agents API picks up computer use, tool search and context compaction.
OpenAI also introduced Ultrafast, a premium tier that pushes token generation in Codex up to eight times faster, to roughly 300 tokens per second. In the API it costs six times the standard rate, or $60 per million input tokens and $300 per million output tokens on GPT-6 Astra.
