thirdbrain.
Book a call

AI basics · Part 3

There is a menu beside the message box. Does it matter which one I pick?

How to choose an AI model and effort level

Two dials: which model, and how hard it thinks. Bigger is not always better.

4 minute read Updated

Download the PDF

Start with whatever model the product gives you by default and leave the thinking setting alone. Move to a faster model for simple, high-volume work. Move to a more capable model, or let it think for longer, when the problem needs careful reasoning. Bigger and slower is not automatically better for every task.

Where the settings live

The menu beside the message box usually holds two separate controls: a model selector, and a setting for thinking or effort. The labels differ between Claude, ChatGPT and Gemini. If you have never changed them, every answer you have had has used whatever the product picked for you.

Microsoft Copilot works differently here. Instead of two separate controls it offers a single combined choice, so you pick one option rather than setting a model and an effort level separately.

Choose by the job, not the model name

Choose an AI model by the difficulty of the job, not by memorising the product name.

Model names and versions change quickly. The description beside the model is more useful than the label. Most model lists fall into three practical groups: fastest answers, everyday work and hardest problems.

Choose the smallest model that can do the job well.

Effort controls how long it works

Turn effort down and the model answers faster. Turn it up and the model spends longer reasoning before it responds.

Higher effort also uses more of your allowance and may cost more in products that charge by usage. Some tools call it thinking, reasoning or deep think.

Same model. Different amount of thinking.

Where to start, and when to change

Start with the everyday model and its normal effort setting, then move down for simple work or up for difficult reasoning.

  1. Start with the everyday model and its normal effort setting.

  2. Move down for simple work where speed matters.

    Summaries, first drafts, extraction and routine questions.

  3. Move up for work that needs careful reasoning.

    Strategy, complex analysis, competing constraints or a high-stakes recommendation.

  4. Compare the result before assuming the bigger model is better.

One job, one chat, one model.

A practical model map

Examples current to 7 September 2026. Availability varies by product and plan.

Fastest answers

Simple work where response time matters most.

  • Claude Haiku 4.5
  • GPT-5.6 Luna
  • Gemini 3 Flash-Lite

Everyday tasks

The best starting point for most work.

  • Claude Sonnet 5
  • GPT-5.6 Terra
  • Gemini 3 Flash

Hardest problems

Complex reasoning, analysis and multi-step work.

  • Claude Opus 5 or Fable 5.1
  • GPT-5.6 Sol or GPT-6 Astra
  • Gemini 3 Pro

OpenAI launched GPT-6 Astra on 3 September 2026. It appears as GPT-6 Pro in ChatGPT. GPT Terra and Luna are available in Work, Codex and OpenAI's developer tools, not standard ChatGPT conversations. Model lists will keep changing.

What each model trades

As models get bigger, answers usually get slower and more expensive, and the work they can handle gets harder.

Model size

Smaller Bigger

How fast it answers

Speed usually falls as the model and effort increase.

Complexity it can handle

Capability usually rises with a more capable model and more thinking time.

How much it costs

Allowance use usually rises with model capability, effort and response length.

These are coarse groups, not scores. On a monthly chat plan, cost often shows up as more of your allowance rather than a separate bill.

Try this in three minutes

Run the same prompt on the fastest model and the most capable model, then decide whether the slower answer earned the extra time.

  1. Open a new chat. Choose the fastest model and lowest effort available, then paste a real prompt.

  2. Open another new chat. Choose the most capable model and highest effort available, then paste the same prompt.

  3. Time both answers and read them side by side.

  4. Decide whether the slower answer earned the extra time and allowance.

What did you notice?

Share this guide

Checked against the source

Product details change. These are the sources.

Quick answers

Common questions about AI models and effort levels.

Which AI model should I use for everyday work?

The everyday model your product already loads by default is built for most day to day work, so start there before changing anything. Drafting, summarising, extraction and routine questions rarely need a bigger or slower option. Only switch model or effort when you can point to a specific reason, such as a genuinely hard problem or a task where speed matters more than depth.

What does effort or thinking level mean?

Effort or thinking level is the setting that controls how long the model reasons before it answers you. Turning it up can improve difficult, multi-step answers, but it also takes longer and uses more of your monthly allowance. Turning it down gives a faster, lighter-touch answer that suits routine work.

Does letting AI think for longer actually give me a better answer?

A longer thinking time helps most on problems with several moving parts, competing constraints or a chain of reasoning steps, not on simple requests. On an easy question, extra thinking time often just adds delay without changing the answer. The only reliable way to tell is to compare the fast and the slow answer on your own real question, side by side.

Is a bigger, more capable AI model always worth choosing?

A bigger AI model is not automatically the better choice for every task. It earns its cost on genuinely difficult work: complex strategy, detailed analysis, conflicting constraints or a high-stakes recommendation where a weak answer would be costly. For routine work, a smaller or faster model usually finishes just as well and much sooner, so running the top tier by default is often paying for capability you are not using.

Does choosing a more capable model or higher effort setting cost more?

Choosing a more capable model or a higher thinking setting almost always uses up your monthly allowance faster. Some usage-based tools also charge more per response for the bigger option. On a fixed monthly plan you will not see a separate bill, but you will hit your limit sooner if you leave the top setting on for everything.

My organisation chose Microsoft Copilot for me. Does any of this still apply?

Some workplace AI tools, including Microsoft Copilot, combine the model and effort choice into a single control that picks the balance for you automatically. You can still ask for a quicker answer or request deeper reasoning when the automatic choice does not match the job, even without seeing a model name to choose from. The skill stays the same either way: notice whether a fast answer was good enough, or whether the task deserved more thinking time.

A first conversation

Want help applying this to your work?

Bring one real task. We will work through it together and sharpen your setup as we go. No deck, no pitch.

Book a free discovery call