OpenAI DevDay 2026: What Changed for Codex, GPT-6 Astra Ultrafast and Developers

What changed for Codex at DevDay 2026?

OpenAI’s DevDay 2026 recap covers more than 20 announcements across ChatGPT, Codex, models, and developer tools. For developers working with Codex, the changes center on four areas:

• Codex Cloud: Start and continue tasks across desktop, web, and mobile using reusable environments.
• CLI workflows: Use voice to start and steer tasks, and use the /agents view to delegate and track multiple tasks.
• Code review: Inspect pull-request summaries, diffs, and potential issues in the ChatGPT desktop app, with automatic reviews running in the cloud.
• Model speed: Access GPT-6 Astra Ultrafast through eligible subscriptions or configure it through the Responses API.

The dates matter because the recap brings together features with different release histories. Voice controls and default worktree support shipped in Codex CLI 0.156.0 on September 22. Version 0.160.0 followed DevDay on October 1 with additional interface changes, session behavior, and fixes.

The API announcements also have separate timelines. The Agents API entered public beta on September 10; computer use was the September 29 addition. Treating every recap item as a new DevDay launch would obscure what actually changed.

For developers comparing these workflows with Claude Code, the Claude Code versus Codex guide provides a place to continue that comparison. Here, the focus is the dated OpenAI announcements and releases available by October 2.

Codex Cloud and code review: work that continues while you are away

Reusable environments across devices

Codex Cloud lets developers start and continue tasks from desktop, web, or mobile. Reusable environments contain repositories, tools, and dependencies, while each task gets an isolated workspace.

A task can continue while the developer’s computer sleeps. That answers a practical question about cloud execution: closing a laptop does not require the cloud task to stop.

Reusable environments also provide shared approved settings and permissions. Managed workspaces retain their cloud-access controls, so the expansion should be read alongside the organization’s existing access rules.

The DevDay recap lists Codex Cloud availability for Plus, Pro, Business, Healthcare, Education, and Enterprise. That availability list is separate from the eligibility rules for Astra Ultrafast: access to a cloud workspace does not establish access to every model speed tier.

When evaluating this workflow, check the repository, tools, dependencies, and approved permissions in the reusable environment before starting work. Those are the components the announcement identifies as the basis for continuing tasks across devices.

Desktop review and automatic cloud review

The Codex code review announcement adds pull-request summaries, diffs, and questions about potential issues inside the ChatGPT desktop app. It covers feedback for GitHub pull requests and GitLab merge requests, and the recap lists availability on all plans.

Automatic cloud reviews can run while the developer is away. The desktop interface and cloud execution therefore address different parts of the review workflow: the app provides a place to inspect changes and discuss potential issues, while the cloud can perform automatic reviews without requiring the developer to remain at the computer.

For a team evaluating the feature, distinguish the review surface from where review work runs. The desktop announcement describes how developers interact with summaries and diffs; the automatic cloud review announcement describes work continuing in their absence.

The Claude Code web and desktop guide is a related reference for readers comparing how they access coding tools across interfaces.

CLI voice, parallel agents, and the 0.160.0 update

Voice and task management were already arriving before DevDay

The refreshed CLI supports starting and steering tasks with voice. Its /agents view supports delegation and tracking multiple tasks, while the recap also highlights improvements to prompt editing, session resumption, worktrees, and terminal readability. The recap lists the refreshed CLI as available on all plans.

Voice has a more specific release date. Codex CLI 0.156.0 enabled voice conversations by default on September 22, added an F8 toggle, and introduced:

/voice settings

That release also bundled Linux and Windows audio runtimes, added /usage analytics, and enabled worktree support by default.

These details establish which controls were available before DevDay. F8 and /voice settings belong to the September 22 release history; they should not be presented as additions introduced by 0.160.0.

For multiple tasks, /agents is the announced view for delegation and tracking. Worktree support is another part of the CLI workflow, but the brief does not establish a complete sequence of commands for creating or managing those worktrees.

What actually changed in Codex CLI 0.160.0?

Codex CLI 0.160.0 shipped on October 1 under the GitHub tag rust-v0.160.0, at commit a956835. To install that specific version:

npm install -g @openai/codex@0.160.0

The release adds keyboard access to “Show more” for older tasks. It also adds fullscreen transcript selection and middle-click paste on supported local Linux X11 terminals.

Session behavior received changes as well. Sessions can run outside projects using workspace defaults where policy permits, and permissions are restored on resume. Opt-in Guardian review capabilities can use earlier instructions and agent-handoff context.

The fixes address several problems that matter during longer sessions: queued messages recover after reconnection without duplicate sends; provider, reasoning-summary, and verbosity settings are preserved; and resume/fork history is corrected. The release also includes Windows sandbox and background-console fixes, plus improvements to subagent environment startup and SQLite initialization.

For an upgrade decision, separate feature needs from reliability needs. Voice and default worktree support were already present in 0.156.0; the October 1 release adds the specific interface and session changes above, alongside fixes for reconnection, settings, history, and environment startup.

GPT-6 Astra Ultrafast: speed claims, plans, and credits

OpenAI describes GPT-6 Astra Ultrafast as a premium speed tier. Its September 29 announcement advertises up to eight times faster token generation in Codex, at 300 tokens per second, and up to six times faster generation in the API.

Those figures describe advertised token-generation speed. They do not establish that a complete coding task, test run, or review finishes eight times faster. Use the figures to assess generation speed, and evaluate complete workflows separately.

Subscription access

The dated September 29 ChatGPT release notes specify that Pro 500 costs $500 per month and includes Astra Ultrafast in ChatGPT Work and Codex. Eligible users select Ultrafast in the model picker.

The distinction between eligibility and usage allowance is central. Credits can cover usage after the included Pro 500 allowance, but buying credits on Pro 100 or Pro 200 does not grant access to Ultrafast.

The announcement also describes GPT-6.1 Sol Ultrafast as “coming soon.” As of this article’s cutoff, that statement should be treated as a future availability announcement.

For related model coverage, see the GPT-6 Astra guide.

API configuration and the other developer announcements

Enable Astra Ultrafast through the Responses API

The September 29 API changelog confirms Astra Ultrafast availability for API customers, subject to rate limits. Use the Responses API with these request fields:

{
  "model": "gpt-6-astra",
  "service_tier": "ultrafast"
}

The model ID remains gpt-6-astra; the service_tier field selects Ultrafast. This fragment shows the announced configuration fields, rather than a complete request with application input.

The dated announcement supports global processing and US data residency. EU and other regional inference residency are unsupported for this tier, making residency requirements a concrete consideration before enabling it.

Exact Astra Ultrafast API prices and numeric tokens-per-minute limits are omitted here because their historical status by October 2 could not be established from the research brief. The dated changelog confirms configuration and rate-limited availability without establishing those numbers.

GPT-6.1 Sol and Agents API chronology

GPT-6.1 Sol was also announced on September 29 with the exact model ID gpt-6.1-sol. For prompts with up to 272K input tokens, its standard API prices per million tokens are $2 for input, $0.10 for cached input, $2.50 for cache write, and $10 for output. These are Sol prices, not Astra Ultrafast prices.

Multi-agent delegation in the Responses API is in beta. Separately, the Agents API public beta launched on September 10, and DevDay added computer use on September 29.

That computer-use addition includes an OpenAI-hosted browser, with website approvals and sign-in handled by the application. Developers evaluating it should account for those application responsibilities when planning the workflow.

Key takeaways

• Codex Cloud supports reusable environments and isolated task workspaces, with tasks continuing while the developer’s computer sleeps.
• Desktop Codex code review covers GitHub pull requests and GitLab merge requests; automatic cloud reviews can run while the developer is away.
• CLI voice controls shipped in 0.156.0 on September 22. Version 0.160.0 shipped on October 1 with interface changes, session improvements, and fixes.
• Astra Ultrafast’s advertised gains concern token generation: up to eight times faster in Codex and up to six times faster in the API.
• Pro 500 includes Astra Ultrafast, while credits on Pro 100 or Pro 200 do not unlock it. API activation uses gpt-6-astra with service_tier: "ultrafast".

FAQ

Can Codex Cloud keep working when my laptop is closed?

Yes. The September 29 announcement states that Codex Cloud tasks can continue while the developer’s computer sleeps. Each task has an isolated workspace, and reusable environments contain repositories, tools, and dependencies.

How do I use voice controls and track parallel tasks in Codex CLI?

Codex CLI 0.156.0 enabled voice conversations by default and added an F8 toggle plus /voice settings. The refreshed CLI’s /agents view supports delegating and tracking multiple tasks.

Which plans include GPT-6 Astra Ultrafast, and can credits unlock it?

The DevDay recap lists Astra Ultrafast access for Pro 500 and Enterprise, and Pro 500 costs $500 per month. Pro 500 usage consumes the included allowance before credits; buying credits on Pro 100 or Pro 200 does not unlock Ultrafast.

How do I enable Astra Ultrafast through the API?

Use the Responses API with model: "gpt-6-astra" and service_tier: "ultrafast". The September 29 changelog confirms availability subject to rate limits, with global processing and US data residency supported and EU or other regional inference residency unsupported.