Which ChatGPT Model Should You Use?
It depends on where you're working, because Chat and Work have different menus. In Chat on Plus or higher, select GPT-6 and choose a thinking level. On Free and Go there's little to decide: use the default, and Think for harder questions. In Work and Codex, start with GPT-6.1 Sol for complex work, drop to GPT-6 Luna for simple tasks, and save GPT-6 Astra for the hardest problems.
Here is a task I'm about to do in ChatGPT, [once / every week / many times a day]: [describe the task and paste one real example] My plan is [Free / Go / Plus / Pro], and I'll run it in [Chat / Work / Codex]. In Chat, my options are GPT-6 at Instant, Medium, High, Extra High or Pro, depending on my plan. In Work and Codex, they are GPT-6 Luna (efficient, for focused tasks), GPT-6.1 Sol (complex work) and GPT-6 Astra (the hardest problems). Recommend the lightest model and reasoning effort likely to do this well. Tell me what could go wrong one step lower, and give me one check I can run on the answer to see whether the lighter setting was good enough.
Try it now
- 1Pick a real task from this week and run it in New chat on your usual model.
- 2Open a second chat and change one setting. In Chat, keep GPT-6 selected in the model menu and move the Thinking slider. In ChatGPT Work or the desktop app, use the control beneath the composer: move the slider toward Smarter or Faster, or open Advanced to pick a specific model and effort.
- 3Send the same prompt word for word, then compare both answers against what you actually needed. If you can't point to a difference, keep the lighter setting.
Use it when
Move up when the task is ambiguous, draws on conflicting evidence, or involves a document too long for the Instant window. Move down when you run the same small job often, such as sorting or extracting. It matters most in Work and Codex, where a heavy default drains your allowance on every message.
Skip it when
Don't change models for a rewrite or a short question, because a heavier model adds wait and burns allowance without a visible gain. If you're on Free or Go, the pricing page lists Astra and Sol as unavailable, so there's nothing to compare.
Why it works
In Chat, OpenAI's Help Center says you pick GPT-6 in the model menu, not a model labeled Sol. GPT-6 Sol powers Instant, Medium, High and Extra High, and Pro uses GPT-6 Pro, powered by Astra. Plus includes Medium and High but not Extra High or Pro. GPT-5.6 Sol stays a separate choice, and if you hit a Thinking limit, ChatGPT may continue with another model.
In Work and Codex, OpenAI's Model selection page ranks the current models from most to least capable as Astra, GPT-6.1 Sol, GPT-6.1 Terra and Luna. The stronger ones use your allowance much faster. In Codex and ChatGPT Work on Plus, OpenAI estimates 5 to 45 local messages per five hours on GPT-6 Astra, 15 to 160 on GPT-6.1 Sol, 35 to 285 on GPT-6.1 Terra, and 350 to 3,000 on GPT-6 Luna. Those are estimates, and weekly limits may also apply. The pricing page lists everyday text chats as unlimited, subject to abuse guardrails, so these counts matter most for Work and Codex.
The mode also changes how much fits. On Go and Plus, the GPT Instant context window is 54K, about 40 pages of input. GPT Reasoning is 256K, about 320 pages. OpenAI's own advice is to compare models on the same inputs and keep the lightest setting that meets your quality bar.
Example
Say you're in Work with 40 customer emails to sort by topic. OpenAI describes Luna as its most efficient model for summarization and extraction, and its Model selection page lists Luna at Low effort for simple data extraction. Then you need a memo on whether to change your pricing, based on what those emails say. That means weighing conflicting evidence, which the same page matches to GPT-6.1 Sol at Extra high effort. Same project, two settings.
Going further
In the desktop app and web Work, the default Power setting depends on your account. OpenAI lists these presets: Luna High, Sol Light, Sol Medium, Astra Light, Astra Medium and Astra Extra High, and some paid plans omit the last one. OpenAI suggests starting Luna at High and Astra at Light, so the strongest model at low effort is a sensible choice. Ultra uses subagents on separate parts of a task. If it's missing, turn on Ultra in model picker slider under Settings > Configuration. Speed costs extra too: Fast uses included usage at 2.5 times the Standard rate, and Ultrafast at 8 times.
OpenAI's pages disagree about Free and Go. The pricing page and the October 7, 2026 release notes say GPT-6 Luna. The Help Center's models article and Free Tier FAQ, both updated around October 30, 2026, say GPT-5.6 Luna, and that Think uses it too, while the pricing page lists GPT-5 Thinking Mini. Trust what your own menu shows.
Common mistakes
- Trusting an old guide. OpenAI's Academy fundamentals page, last updated in May 2026, still describes GPT-5 with Fast, Thinking and Pro options. GPT-5.5 left ChatGPT on October 14, 2026.
- Leaving GPT-6 Astra on for routine work in Codex or Work, where OpenAI's estimate on Plus can be as low as 5 messages in five hours.
- Comparing models on different prompts. Change one thing at a time, or the test tells you nothing.
- Forgetting saved settings. OpenAI says to update scheduled tasks, custom agents and scripts that name a retired model.
Sources
- ChatGPT Learn: Model selection
- ChatGPT Learn: Models
- ChatGPT Learn: Pricing
- ChatGPT pricing
- ChatGPT Learn: What's new
- OpenAI Academy: ChatGPT fundamentals
- GPT-6 and other models in ChatGPT, OpenAI Help Center
- ChatGPT Work and Codex, OpenAI Help Center
- ChatGPT Free Tier FAQ, OpenAI Help Center
- ChatGPT release notes, OpenAI Help Center
Checked 6 November 2026. Products change often, so confirm details with the vendor.