What the New ChatGPT Can Do: Work, Codex, and GPT-5.6 Sol vs Terra vs Luna

The new ChatGPT experience is easier to understand when you split it into two decisions. First, choose the product surface: standard ChatGPT, ChatGPT Work, Codex, or the API. Then choose the model tier behind it, such as GPT-5.6 Sol, Terra, or Luna, if that surface and plan let you select one.

That distinction matters because many users are mixing product names and model names together. Work and Codex are workflow surfaces. Sol, Terra, and Luna are GPT-5.6 capability tiers. They are related, but they are not the same thing.

Quick answer: what can the new ChatGPT do?

• It now spans different work surfaces, not just a single chat window.
• ChatGPT Work is designed for project-based, context-aware work chats.
• Codex is the coding workspace for repository, terminal, and developer-tool workflows.
• GPT-5.6 is a model family with three tiers: Sol, Terra, and Luna.
• Model choice depends on where you are using ChatGPT, because not every surface exposes the same options.

In plain English, the new ChatGPT can help with both general knowledge work and technical execution. A non-technical user might use ChatGPT Work to pull together notes, rewrite a proposal, and keep follow-up questions inside the same project-based context. A developer might use Codex for debugging, code generation, or repository-oriented work with local folders and terminals.

The main thing to remember is this: the answer to “what can ChatGPT do now?” depends on both the surface you are using and the GPT-5.6 tier available inside that surface. Standard ChatGPT, Work, Codex, and the API do not all expose the same model choices.

Definition box: ChatGPT Work, Codex, GPT-5.6, Sol, Terra, and Luna

ChatGPT Work vs Codex vs standard ChatGPT: what is the difference?

Standard ChatGPT is the general conversation surface. It is the right default when you want a normal chat without a project-specific or coding-specific workflow around it.

ChatGPT Work is for project-based, context-aware work. OpenAI’s Help Center says a Work chat can be started from a project and uses project context. That makes it more useful for ongoing work such as drafting, revising, briefing, and follow-up analysis that should all stay tied to one workspace context.

Codex is the coding workflow surface. In the desktop app, users choose ChatGPT or Codex from the top-left menu, and if they choose ChatGPT they can switch between Chat and Work. OpenAI also says Codex remains a separate view, its workflows are unchanged, and its history remains separate from ChatGPT history.

For a writer or analyst, Work is usually the better fit. Example: you are preparing a proposal, collecting scattered notes, rewriting sections, and asking follow-up questions across one project. That is a Work problem, not a repository problem.

For a developer, Codex is usually the better fit. Example: you are debugging a service, reviewing a local repository, and using terminal or developer tools. That is a Codex problem, even if the first brainstorming step happened in Work.

The practical rule is simple: choose the interface by task type first. After that, choose Sol, Terra, or Luna if that surface gives you a model choice.

Comparison table: Work, Codex, standard ChatGPT, Sol, Terra, and Luna at a glance

Item	Best for	Strengths	Limitations	Where available
Standard ChatGPT	General conversation, quick questions, lightweight drafting	Simple interface, easy to start, good for normal chats	Less centered on project-based agentic workflows	ChatGPT
ChatGPT Work	Structured work chats, project context, ongoing knowledge work	Continuity across tasks, project-based context, work-oriented conversations	Not the best fit when repository or terminal work becomes central	ChatGPT desktop app under ChatGPT, then Work
Codex	Coding workspace, repository work, CLI and coding workflow surface	Built for local folders, repositories, terminals, and developer tools	Not selectable on web or mobile	Desktop app as a separate view
GPT-5.6 Sol	Complex analysis, deeper reasoning, longer-horizon tasks	High-end reasoning, flagship tier	May be slower or less cost-efficient than lower tiers depending on use case	Availability varies by surface and plan
GPT-5.6 Terra	Balanced everyday work	Balanced capability for routine business tasks	Not always user-selectable everywhere	Available in Work, Codex, and the API on supported plans
GPT-5.6 Luna	Fast, high-volume, cost-sensitive tasks	Fast and affordable, optimized for cost-sensitive workloads	Less focused on maximum reasoning depth	Available in Work, Codex, and the API on supported plans
Availability note	Model selection planning	Helps avoid choosing the wrong surface first	Not every model is selectable in every product surface	Check current plan and rollout status

Availability can change by plan, account type, and rollout stage, so check current official documentation before making process or budget decisions.

What is ChatGPT Work and what can you do with it?

ChatGPT Work is the work-focused mode inside ChatGPT for structured, project-based conversations. Its biggest difference from a normal chat is that it can start from a project and use project context, so the system is working inside a defined workspace context instead of treating every prompt like a fresh session.

That matters when your work unfolds across multiple steps. Instead of repeating the same background in every prompt, you can keep the briefing, draft, rewrite requests, and follow-up questions tied to the same project-based context.

OpenAI also positions ChatGPT Work as a way to gather context, plan an approach, and take action across tools, files, and desktop apps to create work outputs such as spreadsheets, documents, and slides. For business users, that makes Work closer to a work orchestration surface than a simple chatbot.

A practical example: imagine a product marketer building a launch brief. In one Work project, they can collect scattered notes, turn them into a first draft, ask for alternate messaging, generate an executive summary, and then refine the final document without losing the thread.

Another example: a consultant preparing a client proposal can use Work to summarize materials, rewrite sections for different stakeholders, and keep the entire engagement inside one context instead of managing separate disconnected chats.

Desktop use also matters. OpenAI’s Help Center says Voice in Work is available in the ChatGPT desktop app on macOS and Windows. That can make it easier to move through a work session while staying in the same project context.

If you want a deeper setup path, see how to set up ChatGPT Work with project context.

What is Codex now and when should you use it instead of Work?

Codex is the coding-focused surface in the new ChatGPT ecosystem. OpenAI says it remains a separate view in the desktop app, with separate history from ChatGPT, and it can work with local folders, repositories, terminals, and developer tools.

That makes Codex the better choice when code or repo operations are the center of the task. If you are generating code, tracing bugs, inspecting a repo, or working from a terminal-centric workflow, Codex is the better fit than Work.

Work can still play a role before the coding starts. For example, a technical manager might use Work to outline a migration plan, summarize stakeholder requirements, and prepare a task brief. Once the real task becomes repository review, implementation, or debugging, that workflow naturally belongs in Codex.

A direct example: a developer investigating a failing integration can use Codex to inspect the local project, reason through likely causes, propose code changes, and iterate with terminal feedback. That workflow is much closer to a coding workspace than a general work chat.

Another example: a team lead might start in Work to define acceptance criteria and draft a technical handoff, then switch to Codex once branch structure, files, and execution environment become important.

GPT-5.6 Sol vs Terra vs Luna: the simplest way to understand the model tiers

Sol, Terra, and Luna are not three unrelated products. They are capability tiers inside GPT-5.6.

OpenAI describes Sol as the flagship tier, Terra as the balanced tier for everyday work, and Luna as the fastest and most affordable tier. OpenAI also says the GPT-5.6 number identifies the generation, while Sol, Terra, and Luna are durable capability tiers that can improve on their own cadence.

The easiest way to think about the tradeoff is reasoning depth versus speed and cost sensitivity.

When Sol makes sense

Use Sol when the task is complex, ambiguous, or long-running. This is the tier to consider for deep analysis, multi-step planning, or work where reasoning quality matters more than raw speed.

Best-fit example: evaluating several strategic options for a product launch, each with different risks and constraints.

When Terra makes sense

Use Terra for balanced everyday work. It is the middle-ground choice when you want strong capability without defaulting to the highest-end tier for every task.

Best-fit example: routine business writing, summarization, planning, and work chats that need solid performance but not the deepest possible reasoning.

When Luna makes sense

Use Luna when speed and affordability matter most. OpenAI’s API docs describe Luna as optimized for cost-sensitive, high-volume workloads.

Best-fit example: fast repetitive tasks such as bulk summarization, classification, or high-volume assistive operations where latency and budget matter.

Exact behavior, pricing, limits, and access can change over time, so treat these tiers as positioning guidance rather than fixed guarantees.

Which model should you use? A practical decision checklist

Start with the workflow surface, then narrow down the model tier.

• Use standard ChatGPT if you want a normal conversation without project-specific agentic workflows.
• Use ChatGPT Work if you need project context, structured work chats, or desktop-based work features.
• Use Codex if your workflow is coding-heavy, repository-oriented, or CLI-driven.
• Choose Sol for deeper reasoning or longer-horizon tasks.
• Choose Terra for balanced everyday work when available.
• Choose Luna when speed and lower cost matter more than maximum reasoning depth.

Step-by-step framework

1. Define task type. Is this a normal chat, a project-based work stream, or a coding workspace problem?
2. Check complexity. If the task is complex or long-horizon, start by considering Sol.
3. Check speed needs. If you need fast responses at scale, consider Luna.
4. Check efficiency needs. If the task is everyday work and does not need the highest-end tier, consider Terra.
5. Check availability. You may not be able to select every tier in every ChatGPT surface.

Rule of thumb: surface first, then model tier.

Non-coder path: if you are preparing a proposal with several revisions, choose Work first. Then pick Sol for deeper analysis, Terra for regular drafting, or Luna for fast repetitive edits if available.

Developer path: if you are debugging a repo or working from a terminal, choose Codex first. Then use Sol for harder reasoning, Terra for standard engineering support, or Luna for faster high-volume tasks where cost efficiency matters.

For a more detailed framework, see how to choose between Sol, Terra, and Luna.

Where are Sol, Terra, and Luna available? ChatGPT vs Work vs Codex vs API

Availability is one of the biggest sources of confusion.

Surface	Sol	Terra	Luna
Standard ChatGPT	Available through ChatGPT access to GPT-5.6	Not generally presented as selectable in standard chat	Not generally presented as selectable in standard chat
ChatGPT Work	Available on eligible plans	Available on eligible plans	Available on eligible plans
Codex	Available on eligible plans	Available on eligible plans	Available on eligible plans
OpenAI API	Available	Available	Available

OpenAI says Plus, Pro, Business, and Enterprise users in ChatGPT Work and Codex can choose among GPT-5.6 Sol, Terra, and Luna and set an effort level for each. OpenAI also says Free and Go users in ChatGPT Work and Codex access GPT-5.6 Terra, and developers can access Sol, Terra, and Luna through the API.

The practical takeaway is that standard ChatGPT conversations do not necessarily expose the same model picker behavior as Work, Codex, or the API. Check plan access and current rollout status before you standardize a workflow.

What can the new ChatGPT desktop app do?

The desktop app matters because it brings together multiple work surfaces. In the app, users can choose ChatGPT or Codex, and inside ChatGPT they can switch between Chat and Work.

That structure is useful for real work because it lets you move from a normal conversation to project-based work, or from planning to coding, without treating everything as one undifferentiated chat stream.

OpenAI also says Voice in Work and Codex is available in the ChatGPT desktop app on macOS and Windows, with paired iOS remote access, and it is not available as a standalone experience on web or mobile.

OpenAI further says Voice in Work and Codex can be used to start tasks, check progress, ask questions about agents, and coordinate multiple agents through one conversation. For a manager or analyst, that means desktop voice is not just dictation. It can be part of task coordination inside a work session.

Example: a user can talk through a draft rewrite request in Work while keeping the same project context open on desktop. Another user might brainstorm requirements in a standard chat, then move into a Work session for structured execution or into Codex for implementation.

For more on this setup, see ChatGPT desktop app features and voice support.

Real workflow examples: what the new ChatGPT is actually useful for

1. Project-based writing and research

Best surface: ChatGPT Work
Best model tier: Terra for routine work, Sol for deeper analysis

Use Work when you need continuity across notes, drafts, revisions, and follow-up questions. A strategy team can keep a research brief, draft recommendations, and refine outputs in one project-based context.

2. Coding and debugging

Best surface: Codex
Best model tier: Sol for hard debugging, Terra for general engineering support

Use Codex when the real work involves repositories, local folders, or terminal feedback. This is the clearest fit for implementation-heavy workflows.

3. Summarization across workspace context

Best surface: ChatGPT Work
Best model tier: Terra

Use Work to summarize materials tied to one project, then ask follow-up questions without re-explaining the background every time. This is especially useful for consultants, PMs, and analysts.

4. Fast repetitive tasks

Best surface: Work or API, depending on process
Best model tier: Luna

Use Luna when you need speed and efficiency for repetitive workloads. Common examples include large batches of short summaries, triage-style tasks, or other high-volume support work.

Whatever path you choose, test outputs for quality, speed, and cost fit in your own environment. The best tier on paper is not always the best tier for your process.

How usage, credits, and pricing work across Work and Codex

Usage matters because teams often assume each surface has its own separate budget logic. OpenAI says tasks started through Voice draw from the same shared agentic usage and credit pool as Work and Codex.

That means model selection is not just a technical choice. It is also an operations choice. If a team does a high volume of routine work, Terra or Luna may be more sensible than defaulting to Sol for every task.

For example, a support operations team handling many repetitive internal requests may favor Terra or Luna where available, because those tasks may not need the deepest reasoning tier. By contrast, Sol is easier to justify for high-stakes analysis, difficult planning, or challenging engineering work where stronger reasoning may save time downstream.

Pricing and metering can change over time, so use current official rate-card and product documentation before setting policy.

If pricing is your next question, see Codex pricing and shared credits explained.

Limitations, caveats, and common misunderstandings

First, model naming and product naming can change independently. GPT-5.6 is the model family. Work and Codex are product surfaces. Do not treat them as interchangeable terms.

Second, availability is uneven by design. You should not assume that all GPT-5.6 tiers are selectable in every ChatGPT surface or on every plan.

Third, “better” depends on the task. A common misconception is that ChatGPT Work is just standard ChatGPT with a new label. In practice, Work is useful because it is tied to project context and structured workflows, not just because of the model behind it.

Another misconception is that Sol should always be the default. Sol is the strongest fit for harder reasoning, but many everyday business tasks do not need the highest-end tier. Speed, budget, and task volume matter too.

Finally, OpenAI says Sol, Terra, and Luna are durable capability tiers that can advance on their own cadence. So the balance between tiers may evolve over time even if the family name stays the same.

FAQ

What is ChatGPT Work?

ChatGPT Work is a work-focused mode inside ChatGPT for structured work chats. OpenAI says a Work chat can start from a project and use project context, which helps keep related tasks tied to one workspace context.

It is best for ongoing knowledge work such as drafting, summarizing, planning, and refining outputs across multiple steps.

What is Codex now, and how is it different from ChatGPT Work?

Codex is the coding workspace in the broader ChatGPT ecosystem. OpenAI says it remains a separate view in the desktop app and has separate history from ChatGPT.

The difference is workflow focus. Work is for project-based work chats. Codex is for coding-heavy, repository-oriented, and terminal-centered tasks.

What is the difference between GPT-5.6 Sol, Terra, and Luna?

They are capability tiers within GPT-5.6, not separate standalone products. Sol is the flagship tier for deeper reasoning, Terra is the balanced tier for everyday work, and Luna is the fastest and most affordable tier.

The practical tradeoff is depth versus speed and efficiency. Pick the tier that matches the task rather than assuming the highest tier is always best.

Which GPT-5.6 models are available in standard ChatGPT vs Work vs Codex?

OpenAI says broader GPT-5.6 model choice is available in ChatGPT Work and Codex on eligible plans, while developers can access Sol, Terra, and Luna through the API.

Standard ChatGPT does not necessarily expose Terra and Luna as selectable options in normal conversations, so users should verify current access by plan and surface.

What can you do with the new ChatGPT desktop app?

The desktop app brings together ChatGPT and Codex, and inside ChatGPT you can switch between Chat and Work. That makes it easier to move between normal chat, project-based work, and coding workflows.

OpenAI also says Voice in Work and Codex is available on macOS and Windows, which supports task kickoff, progress checks, and agent coordination in desktop workflows.

How do ChatGPT Work, Codex, and the API share usage and credits?

OpenAI says tasks started through Voice draw from the same shared agentic usage and credit pool as Work and Codex. That matters because teams need to think about surface choice and model choice together.

The API is a separate access path for developers, but it still follows model-specific usage logic. In practice, teams should align workflow type, model tier, and current pricing documentation before scaling usage.

Key takeaways

• Work and Codex are related but built for different workflows.
• GPT-5.6 has three durable tiers: Sol, Terra, and Luna.
• Sol favors reasoning depth; Luna favors speed and affordability.
• Model availability differs across ChatGPT, Work, Codex, and API.
• Choose the surface first, then the model tier.

Recommendation: choose the surface by workflow, then choose the model by reasoning depth, speed, and budget.

See the step-by-step guide to choosing the right ChatGPT surface and GPT-5.6 model for your workflow.

References

• https://openai.com/index/gpt-5-6/
• https://help.openai.com/en/articles/20001275-chatgpt-work-and-codex
• https://openai.com/chatgpt-work/
• https://help.openai.com/en/articles/20001274
• https://developers.openai.com/api/docs/models/gpt-5.6-luna