OpenAI gives Codex reusable cloud environments and voice-controlled tasks

At DevDay 2026, OpenAI announced a set of Codex updates centered on one core shift: development work that persists and travels across devices rather than living in a single isolated session.

• Reusable cloud environments: shared, pre-configured coding setups accessible directly from a laptop, phone, or the cloud.
• Voice-controlled CLI: developers can start and steer coding tasks by speaking commands.
• Automated code review: Codex can review changes and flag potential issues before a developer even opens the code.

The catch: this isn't just a UI refresh. It's a structural change to how Codex environments persist between sessions and devices.

Previously, Codex relied on isolated sandbox environments that didn't carry over between sessions. The new reusable cloud environments change that, giving development teams a shared workspace with approved settings and permissions that persists over time.

OpenAI frames the benefit directly: these environments "help tasks start more quickly and give teams a shared workspace with approved settings and permissions," rather than developers reconfiguring an environment from scratch every time they pick up a task.

That continuity matters practically. A developer can start a task on a work laptop and check on its progress from a phone later, without losing context or needing to rebuild the environment on each device.

The refreshed Codex CLI's headline addition is voice control, letting developers start and direct coding tasks by speaking rather than typing commands. Beyond voice, the update includes a new "/agents" view specifically for tracking and delegating multiple tasks at once.

By the numbers: the refresh also brings smaller but meaningful quality-of-life improvements, including prompt editing, session resumption, worktree support, and a cleaner terminal interface designed specifically for longer working sessions rather than quick one-off commands.

OpenAI also added a dedicated code review experience directly inside ChatGPT's desktop app. Developers can review code summaries, examine specific changes, and consult Codex about potential issues before submitting feedback on GitHub or GitLab.

Why it matters: Codex can also perform an initial automated review entirely on its own, while a developer is away from their computer, meaning some first-pass review work can happen without anyone actively driving it. That shifts part of the review workflow from something a developer initiates to something running continuously in the background.

Alongside the developer-experience updates, OpenAI introduced Codex Security Cloud, a tool built specifically for scanning and hardening code repositories. It can scan entire GitHub repositories either on demand or on a set schedule, with ongoing checks applied to new commits as they land.

What it actually does: Codex investigates any findings, removes duplicate issues, and prepares fixes in the cloud, all of which can happen even with a developer's laptop closed. The toolset also includes access to models from OpenAI's Daybreak Blue cybersecurity initiative, without developers needing to submit a separate application to use them.

That combination, continuous scanning plus automated fix preparation, positions Codex Security Cloud as something closer to an always-on security team member than a tool a developer runs manually when they remember to.

The updated Codex features are available across a wide range of OpenAI's subscription tiers: Plus, Pro, Business, Healthcare, Education, and Enterprise. That breadth suggests OpenAI is treating these updates as broadly relevant infrastructure rather than a premium feature reserved only for its highest-paying enterprise customers.

These Codex updates arrive alongside a new underlying model, GPT-6.1 Sol, which OpenAI says nearly matches its flagship GPT-6 Astra model's coding performance at roughly a fifth of the cost. On DeepSWE v1.1, a benchmark for complex software-engineering tasks in real codebases, GPT-6.1 Sol scored 75.2% at higher reasoning settings, beating GPT-6 Sol's prior best of 68.8%.

Why that matters for Codex specifically: OpenAI reports up to 8x faster token generation in Codex and 6x faster in its API as a direct result of the new model, the kind of speed improvement that makes features like voice-controlled tasks and continuous background security scanning practical rather than frustratingly slow.

The model's gains extend beyond raw coding benchmarks too. On AutomationBench, which tests whether an AI agent can correctly complete multi-step business workflows, GPT-6.1 Sol scored 2.2 percentage points above Anthropic's Opus 5.5 at medium reasoning effort, while costing roughly a third as much per task.

On OSWorld 2.0, a benchmark for computer-use tasks like navigating interfaces and completing on-screen actions, it outperformed GPT-6 Sol by seven percentage points and came within 2.1 percentage points of Astra's score at about one-seventh the cost per task.

Why those specific benchmarks matter for Codex: reusable cloud environments and automated security scanning both depend on an underlying model that can reliably complete multi-step tasks without constant human correction. A faster, cheaper model that still performs close to OpenAI's top-tier Astra is what makes running those features continuously, rather than only during active coding sessions, financially and practically viable at scale.

Codex isn't operating in an uncontested space. The AI coding assistant market has shifted dramatically over the past two years: GitHub Copilot held a near-monopoly after its 2021 launch, Cursor became the dominant agentic coding tool through 2025, and by early 2026, Anthropic's Claude Code had overtaken both in professional usage and developer satisfaction.

Codex's own trajectory has been notably fast. The tool rose from near-zero usage in mid-2025 to more than 3 million weekly active users by April 2026, with one survey of senior engineers finding Codex usage had already reached roughly 60% of Cursor's.

GitHub has since moved to integrate both Claude and Codex directly into Copilot itself, letting developers choose between competing AI coding agents inside a single core development experience rather than picking one tool exclusively.

That competitive backdrop is exactly why these updates matter beyond their individual features. Reusable cloud environments, voice control, and continuous security scanning are Codex's answer to a market where developer loyalty has proven to shift quickly, and where feature parity alone hasn't been enough to hold onto share.

These Codex updates didn't launch in isolation. They arrived the same day as ChatGPT Space, OpenAI's new collaborative workspace product, Dots, its always-on AI agents, and the GPT-6.1 Sol model, all part of a broader slate of more than 20 announcements at DevDay 2026.

Why that context matters: taken together with Space and Dots, these Codex updates reflect a consistent theme running through OpenAI's DevDay announcements this year, building persistent, cross-device, continuously working AI tools rather than session-based assistants that reset every time a user closes the app.