Choosing your AI model: the right tool at the right cost¶

Level: Beginner

Not every AI assistant is equally good at every task. Some answer almost instantly at minimal cost, others take more time and use more resources. This tip gives you a simple method to pick the right tool on the first try.

The everyday problem¶

You open an assistant to rephrase an email and get a heavy, needlessly detailed answer. Or the other way around: you hand a complex file to a lightweight tool and the result stays shallow. Either way, you lose time, resources and trust in the tool.

The principle: economy first, higher tier second¶

Always start with the economy model: it is fast, cheap and good enough for most everyday tasks. Then assess the result. Move to a higher-tier model only if the answer is still not good enough: incomplete reasoning, logic errors, or instructions too complex for the basic tool.

Reading the colored dots¶

Your model picker shows a colored dot next to each model. It tells you its cost level.

The habit to keep: start your task with a 🟢 model. If the answer does not fit, run it again with a 🟡 model. Keep 🔴 for truly complex cases.

Concrete examples¶

• Rephrasing an email (🟢): the economy model is enough. State the tone and length you want.
• Summarizing meeting minutes (🟢): start with the economy tool, then check the key points.
• Analyzing a long document (🟢 then 🟡): try the economy tool first, then move to intermediate if the reasoning stays weak. Anonymize the document beforehand.
• High-stakes file (🔴): keep premium for cases where the first two levels failed.

When should you move up?¶

First, test the economy model with a clear instruction. If the result does not fit, move up a level as soon as you tick at least two boxes:

• The answer is wrong, incomplete or shallow.
• A precise follow-up did not improve the result.
• The task requires reasoning in several steps.
• The context is long or highly technical.

On video¶

Two demos to see the settings in action.

Common mistakes¶

• Picking the 🔴 model by reflex, "just in case", without assessing the real need.
• Judging a tool on a single answer, when a better-worded instruction often changes the result.
• Forgetting to state the context, the target audience and the expected format.
• Not proofreading the answer before using it.

Hands-on exercise (5 minutes)¶

1. Minute 1: write a clear instruction to rephrase an anonymized email.
2. Minute 2: run it with a 🟢 model and rate the result from 1 to 5.
3. Minute 3: run it once more with one extra detail (tone, length, audience).
4. Minute 4: if the result is still not good enough, try a 🟡 model and compare.
5. Minute 5: conclude in one sentence: which level do you keep for this task, and why?

PICOT security reminder¶

Reading time: 4 minutes.