What’s New In ChatGPT’s Latest Models? GPT-5.6 Sol, Terra And Luna Explained

ChatGPT is no longer best understood as one chatbot with one level of intelligence. The latest GPT-5.6 generation is a family of models designed for different kinds of work: a flagship model for demanding reasoning and coding, a balanced option for everyday professional tasks, and a faster lower-cost model for high-volume work.

That shift matters. The practical question is no longer simply “Is ChatGPT good?” It is: which model should handle this task, how much thinking does it need, and when is another AI platform the better tool?

OpenAI’s current family is made up of GPT-5.6 Sol, Terra and Luna. The names are intended to represent durable capability tiers: Sol for the most difficult work, Terra for balanced day-to-day performance, and Luna for speed and scale. For people using ChatGPT Work, Codex or the API, choosing the right tier can make a noticeable difference to quality, speed and cost.

GPT-5.6 Sol, Terra and Luna at a glance

Model	Designed for	Best use cases	Main trade-off
GPT-5.6 Sol	Frontier reasoning and difficult professional work	Complex coding, multi-step research, data analysis, planning, design and agentic workflows	Higher usage cost and more time spent thinking through difficult tasks
GPT-5.6 Terra	Balanced everyday intelligence	Business research, drafting, analysis, practical coding, content workflows and repeatable automations	Less headroom than Sol on the hardest technical or long-horizon work
GPT-5.6 Luna	Fast, cost-efficient scale	Summaries, classification, extraction, first drafts, simple support workflows and high-volume transformations	Not the right model for demanding reasoning or complex autonomous tasks

The big change: models are becoming work systems

Earlier generations of AI were mostly used to answer questions or produce text. Modern AI models increasingly act more like work systems. They can research, inspect files, analyse spreadsheets, create documents, use connected tools, write and debug code, and continue through a multi-step task instead of stopping after a single response.

ChatGPT’s agent capabilities are part of this change. Rather than only suggesting how to complete a task, an agent can use an available toolbox to research, build a presentation, work with files or move through an approved workflow. The important word is approved: capable AI should still have boundaries, permissions and human review around important decisions.

For businesses, this means AI is moving from “help me write an email” to work such as:

• Turning a collection of customer notes into a structured action plan.
• Reviewing competitor websites and extracting useful patterns.
• Creating first-draft reports from uploaded data.
• Generating, reviewing and improving code across a project.
• Summarising lengthy research, contracts or internal documentation.
• Building repeatable internal processes that save time every week.

That does not mean every task needs the strongest model. In fact, choosing a high-end reasoning model for simple summarisation is often wasteful. The real advantage of the Sol, Terra and Luna structure is that it gives people a more sensible choice between depth, speed and cost.

GPT-5.6 Sol: the model for hard problems

GPT-5.6 Sol is OpenAI’s flagship option in this family. It is designed for demanding reasoning, complex coding, long-running tasks and professional work where the quality of the plan matters as much as the final answer.

OpenAI describes Sol as its strongest coding model to date. In its launch materials, the company said: GPT-5.6 Sol is our best coding model yet.

That does not mean Sol is only for developers. It is useful whenever a task has several dependent steps and a weak early assumption could derail the whole outcome. Examples include analysing a large dataset, planning a website migration, comparing complicated business options, investigating why code has failed, or turning messy operational notes into a usable system.

Sol is also more suitable when users need the model to challenge the task rather than blindly follow it. Stronger models can be more useful at identifying missing information, highlighting risks, testing assumptions and suggesting a better route to the goal.

For software teams, the jump is especially important. Coding assistants are no longer limited to writing isolated snippets. They can inspect repositories, trace bugs through several files, propose a plan, make changes and test the outcome. That is one reason the wider AI industry is putting so much focus on coding agents. AskZyro previously explored this shift in its look at SpaceX buying AI coding tool Cursor, and it is a trend likely to reshape how small teams build products.

GPT-5.6 Terra: likely the best everyday choice

Terra is the balanced model in the GPT-5.6 family. It is aimed at the work most people actually do every day: research, planning, content creation, analysis, document handling, practical coding and structured business tasks.

In simple terms, Terra is the model to reach for when you want a strong answer without paying for maximum reasoning on every task. It is designed to offer a more efficient middle ground between frontier capability and high-volume speed.

For a business owner or marketer, Terra is well suited to tasks such as:

• Turning rough notes into a client-ready proposal.
• Researching a market and summarising the important findings.
• Creating content briefs, email sequences and campaign ideas.
• Reviewing a spreadsheet and identifying patterns or anomalies.
• Drafting website copy and then refining it against a brand voice.
• Writing or improving smaller pieces of code, scripts and automations.

It may be tempting to assume that the flagship is always the best choice. In reality, a faster balanced model can be better for everyday productivity because it allows people to iterate quickly. A good workflow is often: use Terra for the first pass, evaluate the output, then use Sol only for the most difficult reasoning, high-stakes review or complex build work.

That is also where AI has become genuinely useful to small teams. The biggest gains usually do not come from one impressive prompt. They come from removing dozens of small bits of friction: drafting, reorganising, researching, formatting, comparing and summarising. For more practical examples, see our guide to how AI assistants save time.

GPT-5.6 Luna: speed and scale for routine work

Luna is the faster, more affordable member of the family. It is not intended to beat Sol on the hardest reasoning tasks. Instead, it is designed for situations where speed, volume and cost control matter more than squeezing out the final few percentage points of quality.

That makes Luna useful for tasks such as classifying support tickets, extracting fields from documents, producing first-pass summaries, converting information into structured formats, generating product-description variations or processing large lists of similar requests.

For example, an agency might use Luna to categorise hundreds of publisher prospects by niche before a human reviews the best opportunities. An ecommerce business could use it to create first drafts of product metadata. A support team could use it to route incoming requests before a person handles anything sensitive or complex.

The key is not to see Luna as “worse” AI. It is optimised for a different job. Using a cost-efficient model for repetitive work can leave more budget and attention for the problems that genuinely require deeper reasoning.

ChatGPT Work: from conversation to connected work

ChatGPT Work is where these newer model capabilities become particularly interesting for teams. Rather than treating AI as a separate website where ideas go in and text comes out, the goal is to connect AI to the work already happening across documents, projects, data and approved tools.

With the right access and human oversight, this can turn ChatGPT into a more practical collaborator. It can help pull together research, work with shared files, create documents, assist with planning and support longer-running tasks.

However, teams should avoid giving any AI system broad access without thinking through permissions. Use the least access needed, require approval for meaningful external actions, review outputs that affect customers or finances, and avoid uploading confidential data unless the organisation’s settings and policies allow it.

Introducing a limited preview of GPT-5.6 Sol, our next generation frontier model, as well as GPT-5.6 Terra, a balanced model for efficient, everyday work, and GPT-5.6 Luna, a fast and affordable model for high-volume work.

— OpenAI (@OpenAI) June 2026

How does ChatGPT compare with Claude and Gemini?

The major AI platforms are converging in some ways. ChatGPT, Claude and Gemini can all help with writing, research, analysis, coding and multimodal tasks. The difference is increasingly about workflow, connected tools, context, reliability on specific tasks and how well the platform fits into the software a team already uses.

Platform	Current direction	Where it can stand out	Best practical use
ChatGPT / GPT-5.6	Three-tier Sol, Terra and Luna model family, with increasing focus on reasoning, coding, tools and work completion	Broad general capability, agentic workflows, coding support and flexible model choice	Teams wanting one adaptable AI workspace for research, documents, analysis and technical work
Claude	Strong emphasis on long-context work, careful reasoning, coding and autonomous tool use	Long documents, detailed writing, large codebases and focused analytical tasks	Users who value sustained context and thoughtful drafting or engineering support
Gemini	Deep integration with Google products, multimodal creation and action-focused AI	Google Workspace workflows, Google Search-adjacent research and multimodal tasks	People and teams already heavily invested in Google’s ecosystem

Claude’s latest releases continue to focus on agentic coding, long-running workflows and handling large amounts of context. Gemini’s current direction centres on multimodal work and its close connection to Google’s ecosystem, from Workspace to Android and Google’s AI tools.

There is no permanent winner. The best platform depends on the task. A team doing extensive work in Google Workspace may naturally find Gemini convenient. A developer working through a large codebase may prefer Claude or GPT-5.6 Sol depending on the project. A business that wants a broad work assistant for research, documents, data and repeatable tasks may find ChatGPT Work especially compelling.

The sensible approach is to test the same real task across two or three platforms. Do not compare them with a vague question. Give each model the same brief, the same data and a clear definition of success. Then assess factual accuracy, usefulness, speed, ability to follow instructions and the amount of human correction required.

AI models are improving, but they still need judgement

More capable AI does not remove the need for human judgement. It changes where human effort is most valuable.

People should still set the goal, provide context, check claims, make final decisions and take responsibility for customer-facing or high-stakes work. AI is excellent at accelerating research, structure, drafts and repetitive tasks. It is less reliable when asked to make unverified claims, interpret ambiguous business context without background information or act without clear boundaries.

This is especially important as AI companies compete to make their systems more autonomous. The story is not just about capability; it is also about safety, access and responsible deployment. Our article on why Anthropic shut down Fable 5 looks at why powerful models can raise new questions around control and safeguards.

What about the environmental cost of more powerful AI?

As AI models become more capable, more people are rightly asking about their energy and water use. There is no single simple number because impact varies by model, data centre, cooling system, electricity source, hardware and the length of each task.

The most useful takeaway is that efficiency matters. Using the appropriate model for a task is not only sensible for speed and cost; it can reduce unnecessary compute. A quick extraction or summary should not require maximum reasoning. A difficult technical problem may justify more compute because it saves meaningful human time or prevents expensive mistakes.

For a fuller look at this topic, read our guide on how much water AI uses.

Which GPT-5.6 model should you use?

Choose GPT-5.6 Sol when the task is difficult, high-value or genuinely multi-step. This is the model for serious coding, deep research, complex planning, technical troubleshooting and high-quality review.

Choose GPT-5.6 Terra for most professional day-to-day work. It is likely the sensible default for research, analysis, business writing, practical coding, documents and repeatable workflows.

Choose GPT-5.6 Luna when you need speed and volume. It is ideal for structured, routine work where a human can check the result and maximum reasoning is unnecessary.

The bottom line

The most important change in the latest ChatGPT models is not simply that they are smarter. It is that users can choose a level of capability that matches the work.

Sol is for the hardest problems. Terra is the practical everyday model. Luna is for fast, high-volume work. Together, they make it easier for people and businesses to use AI more deliberately instead of treating every request as if it deserves the same level of time, cost and reasoning.

That is where AI becomes genuinely useful: not as a novelty, but as a flexible layer of support across research, writing, analysis, coding and operations. The businesses that benefit most will be the ones that choose the right model, keep people in control and build repeatable workflows around real problems. Of course, future models will advance things further, such as ChatGPT6, which is likley to land in the not so distant future. 

Model availability, usage limits and capabilities can vary by plan and may change as providers continue rolling out updates.