OpenAI expands Codex and its API at DevDay with security scans, a Decisions API, and Ultrafast

What happened

OpenAI used its DevDay 2026 event to announce a significant expansion of Codex and its developer API. Codex now supports reusable cloud environments, automatic security scans for connected GitHub repositories, and a dedicated code review view within the desktop application.

On the API side, OpenAI extended its Agents API to support Computer Use, allowing agents to interact directly with software interfaces rather than relying solely on text-based commands. The company also introduced a new Decisions API, designed specifically for fast, single-step decision-making tasks rather than extended multi-turn reasoning.

Alongside these updates, OpenAI launched Ultrafast, a premium processing tier that the company says delivers responses up to eight times faster than standard tiers, at roughly six times the cost.

Why it matters

This is fundamentally a story about what OpenAI is making technically possible for developers and enterprises building on its platform, rather than a customer-facing product shift. The move toward reusable cloud environments and repository-level security scanning signals that Codex is being positioned less as a coding assistant and more as infrastructure for continuous, semi-autonomous software development — a meaningful step in how engineering teams might structure their workflows.

The addition of Computer Use to the Agents API and the introduction of a Decisions API point to a broader architectural shift: OpenAI is differentiating between agents that need to reason over many steps and those that simply need to make one fast, reliable call. For organisations building automated systems — including service and support workflows — this distinction matters, because it offers a more precise (and potentially more cost-efficient) way to match the right model behaviour to the right task.

By the numbers

• Up to 8x faster response times promised by the new Ultrafast tier compared with standard processing
• Roughly 6x the price of standard tiers for access to Ultrafast

The Renascence take

The headline feature here isn't speed — it's the quiet introduction of a pricing lever for latency. Ultrafast effectively turns "how fast should this respond" into a cost decision for engineering and product teams, which has real implications for anyone designing AI-driven customer or employee experiences.

Most coverage will focus on Codex's new security scans and cloud environments, but the more interesting signal is that OpenAI now treats speed itself as a premium, purchasable attribute rather than a fixed system property. For service design, this is a reminder that response latency is a deliberate experience variable, not a technical afterthought — and that "instant" answers will increasingly come with a visible price tag attached. Operators building on these APIs should treat the choice between standard and Ultrafast tiers, and between Decisions and multi-step agents, as a customer-experience decision as much as an engineering one — matching response speed and reasoning depth to what the moment actually requires, rather than defaulting to the fastest or most capable option everywhere.