Book a call
Menu
Field Guide · level 1 · Start here

Which model should you pick, and when should you change it?

20 min · as of Oct 2026
You will be able to

Find the model menu in Claude, ChatGPT and Gemini, start on the default, and know when to move up to a heavier model or down to a quicker one (as of Oct 2026).

Part of Hand it the workTime about 20 minutes, with the try-itChecked 16 Oct 2026

Three things to take away

  1. 1

    A model is the brain you pick. Each app offers a quick one, an everyday one and heavier ones for hard or long work. Heavier is slower and uses up your plan's limit faster.

  2. 2

    Start on the default. Move up only when the answer misses something that matters. Move down when you want speed or want to save your limit.

  3. 3

    Next to the model there is a second dial: how hard it thinks before it answers. Claude calls it effort, ChatGPT and Gemini call it a thinking level. It is the same trade: more thorough, but slower and more of your limit.

In plain words

The names you will hear

New model names arrive every few weeks. These are the ones people talk about in Oct 2026, decoded.

You hearWhat it is
Claude Opus 5.5Anthropic's model for long tasks, coding and knowledge work, released 22 Sep 2026. Anthropic says it "performs at the level of Claude Fable 5.1 on most work" and that it writes clearer business and research documents (Anthropic's own claims)
Claude Fable 5.1Anthropic's most capable public model, released 1 Sep 2026. The heaviest thinker: the slowest, and it uses your limit fastest
GPT-6 AstraOpenAI's most capable model, released 3 Sep 2026. You pick it in Work and Codex; in Chat it powers GPT-6 Pro. It is not a Google product: Google's Project Astra is a research prototype that you cannot pick
GPT-6 Sol and GPT-6.1 SolGPT-6 Sol is what runs when you pick GPT-6 in Chat. GPT-6.1 Sol (29 Sep 2026) comes close to Astra on handed-over tasks and is in Work and Codex, not yet in Chat
Gemini 4 ArgonGoogle's newest frontier model, released 13 Oct 2026, shown as Argon in the Gemini app

If you hear a model name that is not in this lesson, look for it in your own model menu before you plan around it. If it is not there, it is not yours to pick.

Claude, ChatGPT and Gemini · as of Oct 2026

Which model when, for office work (as of Oct 2026)

The jobClaudeChatGPTGemini
A quick question, a short rewrite, sorting a listHaiku 5.5 or Sonnet 5.5GPT-6 at InstantAuto, or Flash-Lite
Everyday writing, a summary, a first deckSonnet 5.5 at its Default effortGPT-6 at MediumFlash
A careful report, a messy spreadsheet, a long documentOpus 5.5GPT-6 at High (Extra High on Pro, Business and Enterprise)Pro, thinking High
Handing over a task with many stepsOpus 5.5, in an ordinary Claude conversationWork on Default, or GPT-6.1 SolTask, with Flash or Argon
The hardest, high-stakes workFable 5.1 (uses your limit fastest)GPT-6 at Pro (GPT-6 Pro, powered by Astra), or Astra in WorkArgon, thinking High
As of Oct 2026 · read 16 Oct 2026

In your assistant

Plans decide which models you see. On a work account, your company's plan and admin settings decide. If a model in this lesson is missing or locked in your menu, that is the reason.

Claude

How to pick: click the model name next to the send button and choose a model. Click "More models" to see the rest.

How hard it thinks: in the same menu, click "Effort" and choose Low, Medium, High, Extra high or Max. Each model marks its recommended level "Default". Higher effort gives more thorough answers but uses more of your limit. Thinking stays on for all four current models, so there is nothing else to switch.

ModelUse it forPlans in the Claude app
Claude Haiku 5.5The quick one: summaries, sorting, pulling out details, lots of small jobsEvery plan, including Free
Claude Sonnet 5.5The everyday one: well-scoped tasks, polished documents, slides and spreadsheets. Anthropic: "If you're not sure which model to pick, start here"Every plan, including Free
Claude Opus 5.5Hard problems and long tasksPro, Max, Team, Enterprise (not Free)
Claude Fable 5.1Large, long or high-stakes projectsMax (where it can use up to half of your weekly limit), Team Premium seats, Enterprise; on Pro only through usage credits; not on Free

Older models, such as Fable 5 and Opus 5, are now listed as legacy.

ChatGPT

ChatGPT has two sets of models: one for Chat (ordinary conversations) and one for Work and Codex (the agent and the coding tool).

In Chat: choose GPT-6 in the model menu, then move the Thinking slider.

Slider settingWhat runsPlans
Instant, Medium or HighGPT-6 SolPlus, Pro, Business, Enterprise
Extra HighGPT-6 Sol at its highest thinking levelPro, Business, Enterprise (not Plus)
ProGPT-6 Pro, powered by GPT-6 Astra, for the hardest questions, with its own usage limitPro, Business, Enterprise
  • On Free and Go, the default is GPT-6 Luna, with a Think button for harder questions.
  • On Plus and Pro, Instant no longer moves itself up to a higher thinking level. You choose the level yourself.
  • The older GPT-5.6 Sol can still be picked on Plus and above.
  • Visuals: GPT-6 in Chat can answer with pictures and small interactive tools. To see fewer of them, go to Settings and turn off "Layout and visuals".

In Work: the model picker offers Default (one slider that mixes models and thinking levels for you) or a specific model with its own reasoning levels.

ModelUse it forPlans
GPT-6 AstraThe most capable: using a computer, browsing, software and professional work; OpenAI calls it its "best model for adhering to existing templates" for slides and documentsPlus (limited), Pro, Business Premium seats (full), Business Standard (limited)
GPT-6.1 SolClose to Astra on handed-over tasks, using a computer and professional workPlus, Pro, Business, Enterprise, Edu
GPT-6 Sol, GPT-6 Luna, GPT-5.6 Terra, GPT-5.6 LunaOther choices: Terra balances ability, speed and cost; Luna is the fastestPlus and above; Free and Go get limited Terra in Work on the desktop app

Gemini

The Gemini app shows four simple labels instead of version numbers.

LabelUse it forPlans (personal accounts)
Flash-LiteFast answers to everyday questions; the default without a planEveryone
FlashFast and capable for standard chats and tasksGoogle AI Plus, Pro, Ultra
ProAdvanced reasoning for complex requestsGoogle AI Plus, Pro, Ultra
ArgonGemini 4 Argon, Google's newest, "built to take your most complex tasks from start to finish"Google AI Pro and Ultra

How to pick: at the bottom of the chat box, click the current model and choose another. Locked ones need a higher plan. Then pick a thinking level: Low, Medium or High (Low if you choose nothing). With "auto model selection" on, Gemini picks the model for you.

  • Deep Think is a special reasoning mode for Ultra subscribers, picked from the prompt bar. Its answers take a few minutes.
  • Plans differ by page: Google's plans page lists only Flash for AI Plus, while its help page lists Flash-Lite, Flash and Pro. Check your own menu.
  • Work and school accounts: the models depend on your company's Google Workspace licence.
Start to end

The two-minute test

Anthropic suggests a simple way to choose. Run the same task on a lighter model and a heavier one, and check whether the lighter answer misses anything that matters. If it does not, use the lighter one for that kind of job from now on. If it does, you have found a job that earns the heavier model.

On your own work · nothing private

Try it (10 minutes)

Use something public or made up. Nothing private goes in.

  1. Find the menu (2 minutes). In your main tool, open the model menu and write down the default model and thinking or effort level.
  2. Pick one small job (1 minute), such as turning five made-up bullet points into a short team update.
  3. Run it three ways (5 minutes): on the default, one step lighter, and one step heavier (a different model, or a higher thinking or effort level). Note how long each took and whether anything important differs.
  4. Write your own rule (2 minutes), in three lines: "Quick jobs: ___. Everyday jobs: ___. Hard or long jobs: ___."
Three questions · pick one answer each

Quick check

  1. 1You are not sure which model to use for an everyday summary of a report. What do you do?

    The answer: C Anthropic's and Google's own advice starts from the default. Heavier is slower and uses your limit faster, so let the job earn the step up.

  2. 2A colleague says Astra is Google's new model. Is that right?

    The answer: A In Gemini, Google's newest model is Gemini 4 Argon.

  3. 3On ChatGPT Plus, you leave GPT-6 on Instant and ask for a careful look at a messy spreadsheet. Will it move up to a higher thinking level by itself?

    The answer: B OpenAI changed this in Sep 2026: Instant stays Instant until you move the slider.
0 of 3 answered
Read the whole lessonThe idea in full, the worked examples, common mistakes and what to check

The idea

Think of the models in one app as colleagues who work at different speeds. One answers a quick question in seconds. One handles most everyday work well. One takes its time on the long, careful jobs. You would not ask your most senior expert to sort a list, and you would not hand a junior colleague a board paper on day one.

Every app also has a limit: how much you can use it in a day or a week on your plan. Heavier models and higher thinking levels use more of it. That is the real cost of always picking the biggest model.

So the habit is simple. Leave it on the default. Notice when an answer misses something important. Then move one step up, not straight to the top.

Common mistakes

  • Always picking the heaviest model "to be safe". It is slower, and the heaviest models, such as Fable 5.1, use your limit fastest, often on jobs a lighter model does just as well.
  • Never leaving the quick model, then deciding AI is "not good enough" for hard work.
  • Judging a tool by its free default. Flash-Lite and GPT-6 Luna are built for speed. Compare like with like.
  • Forgetting the second dial. Sometimes the fix is a higher effort or thinking level, not a different model.
  • Planning around a model name you heard but cannot see in your menu.
  • Thinking a heavier model means you can skip the check. It makes fewer mistakes, not none.

What to check before you trust it

  • Did the answer miss anything that matters? That, not the model's name, is the reason to move up.
  • Is the heavier model really better on your task? Run the two-minute test before you make it your habit.
  • Is the model on your plan? If it is locked, ask whether your company's plan includes it before you plan work around it.
  • Whatever the model: check names, numbers and dates against your sources before you send (lesson: How do you check an AI answer before you use it?).
Checked against · read 16 Oct 2026Every claim, its source and the date read

As of: product steps as they were on the vendors' own pages on 16 Oct 2026.

Every product line in this lesson comes from Neosaipien's platform reference (checked on the vendors' own pages, 16 Oct 2026).1. Its sources were each read on 16 Oct 2026. The vendor pages behind this lesson:

Where next

Next lesson · level 2 · 20 min

How do you brief AI so the first answer is close?

Write a request with the goal, who it is for, the source to use and what good looks like, then improve it in one or two turns.

Open the next lesson
The kit · free download

The 2026 AI Starter (free). Its first week starts with finding the model menu and leaving it on the default.

Bring this to your team

The Team workshop.

Book a 20-minute call

Back to level 1, Start here