EN

GPT-6 in ChatGPT: The 3 Models and When to Use Each

#AI Tools
2026-10-12 14:2412 min read

Content Introduction
Ask Questions

The video explains how to choose the right ChatGPT model and effort setting instead of relying on the default option. It describes the three main model tiers—Luna (small/fast/cheap), Soul (mid-range), and Astra (largest/most powerful/most expensive)—and maps them to use cases. Luna is recommended for simple, repetitive, high-volume tasks; Soul for most everyday intellectual work and medium-to-high complexity; and Astra for ambiguous, high-stakes, or highly complex tasks where the best outcome is unclear. The speaker emphasizes using the Work area over Chat for better file access and task handling, selecting the latest model manually, and avoiding overly high effort settings unless truly necessary because they cost more and may not improve results.

Key Information

  • The video explains that ChatGPT has multiple hidden model options behind the scenes, and users should not rely on the default setting because it may select the cheapest or fastest model rather than the most suitable one.
  • It distinguishes between the Chat and Work areas, recommending the Work area for better access to files, tasks, and more advanced functionality.
  • The interface includes effort/speed controls such as instant, medium, high, extra high, and pro, and the user should manually choose the latest model rather than leaving the default.
  • The main model family is described as three key models: Luna as the small model, Soul as the medium model, and Astra as the large model.
  • Most users are advised to stay in the medium-to-high effort range, with medium handling simpler one-input/one-output tasks and high handling larger, multi-source, higher-responsibility tasks.
  • Astra is presented as the strongest but most expensive option, best used when the task is complex, uncertain, or requires exploration of unclear outputs.
  • Luna is recommended for repetitive, highly specific, low-complexity, high-volume tasks such as renaming files, extracting fields from invoices, or sorting large numbers of comments.
  • The speaker emphasizes that in the newer GPT-6 generation, effort level matters more than old prompt tricks, because the model now follows the selected effort setting more consistently.
  • For advanced workflows, the video suggests using subagents in the Work area, with one model orchestrating and others executing or validating tasks in parallel.
  • The overall takeaway is to choose the right model and effort level deliberately, start with the smallest useful setting, and only increase complexity or cost when the task truly requires it.

Timeline Analysis

Content Keywords

ChatGPT model selection

Explains that ChatGPT does not automatically tell users which model is best, so they must manually choose the right model and effort level to improve response quality, speed, and cost efficiency.

GPT-6 effort levels

Breaks GPT-6 into three main effort tiers—instant, medium, and high—and notes additional higher tiers like extra high and pro for eligible plans. The video stresses choosing the right effort level for the task instead of relying on defaults.

GPT-6 Luna

Describes Luna as the smallest and cheapest model, best for simple, repetitive, high-volume tasks such as renaming files, extracting invoice fields, or sorting large sets of comments. It is recommended for lightweight automation and subagent execution.

GPT-6.1 Soul

Presents Soul as the mid-size model and the default choice for most work. It is recommended for intermediate tasks like summarizing a document into a slide, cleaning spreadsheets, and handling multi-step work with moderate complexity.

GPT-6 Astra

Positions Astra as the largest and most powerful model, suitable for complex, high-stakes tasks with vague goals or uncertain outcomes. The speaker advises using it when the desired result is unclear and more reasoning is needed.

Chat vs Work area

Highlights that the Work area is preferable to Chat because it can access files, run more advanced workflows, and support subagents. The speaker recommends using Work whenever possible for better results.

Subagents and orchestration

Shows how the Work area can split a job across parallel subagents, often using Luna for execution and Soul for validation. This setup is useful for large-scale tasks like processing many invoices or accounts efficiently.

Plan limits and pricing

Explains that access to models and effort levels depends on the user’s subscription. Plus users typically get instant, medium, and high, while Pro, Business, and Enterprise plans unlock extra high and pro tiers.

Related questions&answers

What is the main issue with ChatGPT's default model selection?

The default setting often hides which specific model is being used, so users may not realize they are using a smaller, cheaper, faster model instead of the best one for their task.

What are the three main models mentioned in the video?

The three main models are Luna, Sol, and Astra, which correspond to the small, medium, and large model sizes.

Which ChatGPT workspace should most users use?

The work area is recommended for most users because it has access to files and supports more advanced tasks than the chat area.

What does the effort slider control?

The effort slider controls how much reasoning the model uses, and it also affects response quality, speed, and token usage.

Should users leave ChatGPT on the default setting?

No. The video recommends selecting a specific model instead of leaving the default option, because the default may change depending on the effort level and not always use the best model.

When should the Instant mode be used?

Instant mode is best for quick, simple questions when you need a fast answer, especially on the go.

What kind of tasks are best for the Medium level?

Medium is best for one-input, one-output tasks, such as summarizing a document or creating a simple report.

When should the High level be used?

High is recommended for larger or more complex tasks, especially when multiple sources or files are involved, or when the task has higher responsibility.

When should Astra be used?

Astra should be used for highly complex, uncertain, or advanced tasks where you are not sure what the final result should look like.

Why does the video advise against using Pro or Ultra too often?

Because they can cost three to five times more, and the results may not be better than the High level.

What is a good use case for Luna?

Luna is good for repetitive, structured tasks such as renaming many files or extracting specific fields from invoices.

What is the recommended starting point for most users?

Most users should start with the intermediate model and medium effort, then move up only if needed.

How should users choose a model for work tasks?

They should choose a specific model manually rather than relying on the default, and adjust effort based on task complexity.

What happens if you raise the effort level too high?

The AI can become overfocused, spend more time rethinking things, and may produce similar or worse results at much higher cost.

Why is it important to start a new chat for a harder task?

Starting a new chat with the same prompt and context can help when the previous effort level or model didn't produce a good result.

More video recommendations

Share to: