When I first looked at the new GPT-5.6 settings in ChatGPT Work, I genuinely felt that something designed to make work easier had suddenly given me another decision to make.
First, I had to choose between Sol, Terra and Luna. Then I had to select an intelligence level such as Light, Medium, High, Extra High or Ultra. After that, there was another choice between Standard and Faster speed.
That is a lot of configuration before I have even explained what I want to get done.
I work across website development, design, digital marketing, SEO, content, business strategy and client management. My tasks can range from correcting a short email to reviewing a complicated Shopify issue or creating an entire business document. I do not want to spend time choosing an AI stack every time I start a new task.
So I went through the available information and simplified the entire system for myself. This article explains what each setting actually means, how I am choosing between them and what current usage limits look like for ChatGPT Plus users.
This information is based on OpenAI’s published documentation available on July 18, 2026. Models, limits and credit rules can change, so verify the latest information on the official Codex pricing page if you are reading this later.
Why the GPT-5.6 Settings Feel More Complicated Than They Should
The main problem is not that there are too many models. The problem is that ChatGPT presents three different decisions at the same time:
- Which model do you want?
- How much intelligence or reasoning should it use?
- How quickly do you want it to run?
These settings control different parts of the experience. Once I understood that, the system became much easier.
A simple way to think about it is:
- The model determines the type of brain doing the work.
- The intelligence level determines how much time and reasoning that brain should use.
- The speed setting determines how quickly the processing happens and how quickly credits are consumed.
You do not need to optimize all three settings for every prompt. In fact, most people will probably be better off choosing one everyday setup and changing it only when the task clearly requires something different.
GPT-5.6 Sol vs Terra vs Luna
Sol, Terra and Luna are not simply good, medium and bad versions of GPT-5.6. They are designed for different types of work.
GPT-5.6 Sol: For Important, Complex and Open-Ended Work
Sol is the most capable and polished option in the GPT-5.6 family. It is designed for difficult tasks that require deeper judgment, planning, analysis or refinement.
I would choose Sol for work such as:
- Deep research involving multiple sources
- Complex website or application development
- Diagnosing difficult technical problems
- Creating a complete brand strategy or brand book
- Writing end-to-end website copy
- Reviewing important business decisions
- Producing polished documents for clients or stakeholders
Sol is particularly useful when I know the output needs careful thought rather than just a quick answer.
However, using the most capable model for every tiny task is not automatically the best decision. If I only need to correct a paragraph or make a simple CSS change, Sol may be more than necessary.
GPT-5.6 Terra: The Practical Everyday Model
Terra is the everyday all-rounder. It offers strong reasoning and tool use without requiring the full depth of Sol for every request.
This is the model I would use for most normal work, including:
- Shopify and WordPress changes
- Writing or improving client emails
- Marketing copy and social media content
- SEO recommendations
- Meeting summaries and action items
- Website content improvements
- Routine troubleshooting
- Business explanations and planning
For someone like me, whose day moves between development, marketing, content and client work, Terra is probably the most practical default.
It should be capable enough for serious everyday work while helping the available usage last longer than it would with Sol.
GPT-5.6 Luna: For Clear and Repetitive Tasks
Luna is the lighter and more affordable GPT-5.6 option. It is designed for tasks where the instructions are clear and the expected result is already defined.
Examples include:
- Generating multiple meta descriptions from an approved format
- Extracting information from documents
- Categorizing leads, products or content
- Reformatting text into a fixed structure
- Creating several variations based on one approved example
- Summarizing large amounts of straightforward information
Luna should not be treated as a useless model. For repetitive and high-volume work, it may actually be the most sensible choice. There is no reason to use the most expensive brain when the task is already clear and repeatable.
What Do the Intelligence Levels Mean?
After choosing a model, ChatGPT asks how much intelligence or reasoning effort it should use.
The easiest approach is to use the lowest level that consistently produces the quality you need.
Light
Light works best for quick and tightly scoped requests. I would use it for grammar corrections, short rewrites, simple formatting, basic extraction and small code changes where the required outcome is obvious.
Medium
Medium is the balanced setting. It gives the model more room to plan and check its response without turning every task into a deep research project.
This is the intelligence level I would keep selected for everyday work.
High
High is useful when a task has multiple steps, tradeoffs or possible failure points. It is suitable for debugging, strategic recommendations, technical implementation and work that needs more careful verification.
Extra High
Extra High should be reserved for genuinely difficult projects. This includes long documents, multiple sources, complex technical problems, important strategic decisions and situations where missing a detail could materially affect the result.
Using Extra High for ordinary work will usually increase the waiting time and usage without creating a meaningful improvement.
Ultra
Ultra is not simply another step above Extra High. It can divide a large task among multiple subagents, allow them to work on separate parts and then combine their findings.
This can be useful for a project that naturally contains several independent workstreams. For example, one agent could analyze a website, another could study competitors and another could review positioning before the final answer is created.
Most daily tasks do not need Ultra. On a Plus plan, I would use it very selectively because multi-agent work can consume the available allowance much faster.
Standard vs Faster: Is Faster Worth It?
The Faster setting sounds like an obvious upgrade, but there is an important tradeoff.
According to OpenAI’s current documentation, Faster mode increases supported model speed by approximately 1.5 times. However, GPT-5.6 consumes credits at 2.5 times the Standard rate when Faster mode is enabled.
In simple terms, I may receive the answer sooner, but I can complete considerably less total work before reaching my limit.
For that reason, I would keep Standard selected almost all the time. Faster is useful when something is genuinely urgent and waiting is more expensive than the additional credit consumption. It is not something I would leave enabled permanently.
The Three GPT-5.6 Setups I Am Using
I do not want to choose from dozens of combinations every time I open ChatGPT. I have reduced everything to three practical setups.
| Type of work | Model | Intelligence | Speed |
|---|---|---|---|
| Everyday work | Terra | Medium | Standard |
| Important or complicated work | Sol | High | Standard |
| Repetitive or bulk work | Luna | Light or Medium | Standard |
That is the complete system.
For approximately 80% of my work, Terra with Medium intelligence and Standard speed should be enough. I would move to Sol and High only when the task clearly deserves deeper analysis. I would use Luna when the task is repetitive, structured or high-volume.
If someone wants a single setting for absolutely everything, OpenAI’s general Power default currently uses Sol with Medium reasoning. That is also a safe choice. My Terra recommendation is more specific to my workflow, where many tasks are practical, clearly scoped and do not need the full depth of Sol.
How Long Can You Use Each GPT-5.6 Model on ChatGPT Plus?
This is where the answer becomes less straightforward.
There is no honest way to convert ChatGPT usage into a fixed number of working hours. A short email correction and a request involving multiple PDFs, website research, code execution and a long document are both messages, but they consume very different amounts of the allowance.
OpenAI currently publishes the following estimated local-message ranges for Plus users within a shared five-hour window:
| Model | Published local messages per five hours |
|---|---|
| GPT-5.6 Sol | 15 to 90 |
| GPT-5.6 Terra | 20 to 110 |
| GPT-5.6 Luna | 50 to 280 |
OpenAI also states that additional weekly limits may apply, but it does not publish one fixed weekly message number. Usage is now calculated using tokens and credits, so the actual consumption depends on factors such as:
- The amount of context in the conversation
- The size of attached files
- The length of the requested output
- The selected model
- Reasoning effort
- Web research and retrieval
- Tools and code execution
- Prompt caching
- Faster mode
This explains why two users selecting the same model may reach their limits at very different times.
As a conservative planning estimate, I would expect Terra with Medium intelligence to provide around 30 to 70 normal requests in a five-hour window. Sol with High intelligence may provide approximately 15 to 40 substantial requests, especially when working with files and tools. Luna could handle around 80 to 200 lighter requests when the tasks are clearly defined.
These are my planning ranges, not limits officially guaranteed by OpenAI.
How I Plan to Make My Weekly Usage Last Longer
The goal is not to use the weakest model possible. The goal is to avoid spending the highest amount of reasoning and credits on work that does not need it.
Here is the approach I am following:
- I will keep Terra, Medium and Standard as my everyday default.
- I will switch to Sol only for important or complicated tasks.
- I will use Luna for repetitive and bulk work.
- I will avoid leaving Extra High, Ultra or Faster enabled permanently.
- I will provide only the files and context that are relevant to the task.
- I will clearly state the required output, length and boundaries.
- I will start a fresh conversation when an existing chat contains too much irrelevant history.
- I will check Settings > Usage regularly to understand my actual consumption pattern.
The usage dashboard is the best way to calculate personal capacity. If one normal day consumes 20% of the weekly allowance, that usage pattern should support approximately five similar working days. If one major project consumes 40%, I immediately know that I need to use lighter settings for the remaining work.
My Final Opinion
GPT-5.6 is powerful, and I like the fact that we now have more control over quality, reasoning, speed and usage. But the interface currently asks users to understand too many technical choices before they can start working.
I believe the experience would be much more user-friendly if ChatGPT simply offered four clear presets:
- Quick Task
- Everyday Work
- Deep Work
- Parallel Project
The exact model, reasoning level and speed could remain available under Advanced settings for people who want more control.
Until that happens, I am keeping my own system simple:
- Terra Medium for normal work
- Sol High for deep work
- Luna Light or Medium for repetitive work
- Standard speed almost always
AI is supposed to reduce cognitive load, not create another configuration problem. The best setup is not the one with the highest intelligence level. It is the lowest-cost setup that reliably produces the quality the task requires.
That is the rule I am going to follow. Anyway until next time. Stat Strong and Keep Learning.
– Amar Karthik
