The Signal: ChatGPT, Codex, and the GPT-5.6 Workbench

The biggest OpenAI update this month is not one model. It is the workbench around the model. On July 9, the standalone Codex experience moved into the new ChatGPT desktop app for Windows and macOS. Chat, Work, and Codex now sit in one app, while GPT-5.6 Sol, Terra, and Luna give people clearer choices between depth, balance, and speed.

For normal users, the practical shift is simple: ChatGPT is becoming a place where work continues across files, apps, repositories, computers, and phones. For Codex users, Ultra can split a complex job across subagents, the desktop app can edit diffs inline and review pull requests in a side panel, and the CLI can show usage or surface an eligible rate-limit reset through /usage.

This guide sticks to OpenAI documentation checked July 16, 2026. Availability can depend on plan, account, workspace, model, and rollout. If a switch or reset does not appear for you, that is not proof that your installation is broken.

Quick picks

Most people: Start with Power / GPT-5.6 Sol. OpenAI recommends the default Power setting with Sol at medium reasoning for demanding work.
Everyday volume: Use Terra or Luna. Terra is the balanced workhorse; Luna is the faster, lower-cost option for clear and repeatable tasks.
Hard parallel job: Use Ultra selectively. Ultra delegates meaningful parts of a complex task to subagents. It is not needed for routine prompts.
Limits and resets: Check /usage. The Codex CLI can show daily, weekly, or cumulative activity and offer a rate-limit reset when your account is eligible.

Codex moved into the ChatGPT desktop app

OpenAI says that on July 9 the Codex app merged into the new ChatGPT desktop app for macOS and Windows. Codex keeps a dedicated coding experience beside Chat and Work instead of disappearing into a generic chat screen.

The new desktop experience adds inline editing in diffs, pull-request review in a side panel, faster Computer Use powered by GPT-5.6, and multi-repository projects. Existing Codex app users can update normally, choose Codex as the default view, and even keep the Codex logo as the app icon. Desktop Codex projects are also available from the ChatGPT mobile app. OpenAI says the updated desktop app is available globally on every ChatGPT plan, including Free.

Chat, Work, and Codex share one desktop app.
Codex remains a distinct workspace for repository work.
Review and edit diffs without leaving the app.
Group work that spans more than one repository.
Open desktop Codex projects from ChatGPT mobile.

Sol, Terra, and Luna are three different tools

The GPT-5.6 family is not a simple good-better-best ladder. OpenAI describes Sol as the flagship for complex coding, computer use, research, cybersecurity, judgment, and polish. Terra is the balanced everyday model, competitive with GPT-5.5 at lower cost. Luna is the fastest and most affordable member of the family for clear, repeatable work.

In the Codex app, OpenAI recommends starting with the Power setting, which uses GPT-5.6 Sol with medium reasoning. Move toward Smarter when depth matters, Faster when cost or latency matters, and Advanced when you need a particular model or reasoning level.

Sol: ambiguous, difficult, high-value work that needs judgment and follow-through.
Terra: everyday agent work, codebase exploration, review, and balanced tool use.
Luna: extraction, classification, transformation, structured summaries, and other high-volume jobs with a clear target.

Ultra is a team, not just a longer think

Max and Ultra solve different problems. Max gives one selected model more time to reason about a hard task. Ultra uses subagents to handle separate parts of a complex task in parallel, then consolidates the work.

Use Ultra when the task has real independent branches: audit accessibility, verify facts, review code, test mobile layout, and reconcile the results. Do not spend Ultra on a one-paragraph rewrite or a small bug. If Ultra is available to your account but hidden from the desktop slider, OpenAI documents a toggle under Settings, Configuration, Ultra in model picker slider.

Usage, credits, limits, and resets

ChatGPT Work and Codex share pricing, credits, and usage limits. That means work in one surface can affect what remains in another. The exact allowance depends on plan and workload; a long agent run can consume much more than a short edit, so a universal number would be misleading.

In the Codex CLI, /status shows the active model, configuration, context, and rate-limit information. /usage shows account token activity across daily, weekly, or cumulative views and can offer a rate-limit reset when the account is eligible. A reset is an account capability, not a promise that every user receives one on the same schedule. ChatGPT credits can extend usage on supported plans and models.

Run /status before a long job so you know the active model and remaining limits.
Run /usage to understand daily and weekly consumption.
Use a reset only when the interface actually offers it to your account.
Move repetitive support work to Terra or Luna instead of spending Sol or Ultra everywhere.
Higher reasoning, Fast mode, and parallel agents can consume usage faster.

A practical way to choose today

Use Sol at medium for a difficult task you care about. Use Terra for the supporting scans and everyday implementation. Use Luna when the job is narrow and repeatable. Use Max when one problem needs deeper thought. Use Ultra only when the work genuinely splits into separate streams.

The model picker matters less than the brief. State the outcome, point to the files or apps involved, define what done means, and name the checks. A precise task on Terra can beat a vague task on Sol Ultra.

What this means for people who do not code

The desktop merger is not only for software developers. The same app can research, work across documents, use connected apps, operate supported desktop software, and build small tools. Codex is increasingly useful for any job with files, repeatable steps, and a result that can be checked.

Try one bounded workflow: organize a folder of documents, compare a set of quotes, turn notes into a small tracker, or build a one-page calculator. Keep sensitive information out unless you understand the account and data controls, and review the result before relying on it.

Related Power of AI pages

Codex Best Practices: How to brief, supervise, and verify a coding agent.
Agent Loops: How to keep an agent working until a check passes.
The Power of AI in 10 Minutes: A beginner path for trying one useful workflow.
AI in Canada: Canadian privacy, business, school, and local-use starting points.

Sources and official references

Related Power of AI pages

Keep reading with AI Finder, Prompt Studio, ChatGPT vs Claude vs Gemini, the AI glossary, and Which AI Should You Use?.