Advanced · Critical thinking and your own skills · module 128 of 155
Choosing the right model for the task
Choose between fast and reasoning models based on your task, costs, waiting time and your own test.
After this lesson
You compare models using the same test tasks of your own and make a conscious choice based on quality, speed, costs and data agreements.
In view · Along the same yardstick
Choosing a model per task
Fast models for simple tasks, reasoning models for complex analyses.
The same tasks and criteria for each model, including the time spent checking.
Data agreements, a permitted environment and a fixed version for fixed processes.
Part 1 of 3 · Fast or reasoning
Your preference for less movement is on. Choose each part yourself.
Not every model suits every task
AI services often offer several models or modes. Fast, lighter models are suitable for simple tasks such as rewriting or short summaries. So-called reasoning models take more time to work through a problem in steps (lesson 72). They often help with complex analyses, calculations or schedules, but are slower and can be more expensive per question (lesson 80). Model names and capabilities change quickly: what is best today may be different in a few months.
Test with your own tasks
Rankings and benchmarks say something about general performance, but not about your work (lesson 81). So take three to five typical tasks from your test set (lesson 113), have two or three models carry them out and assess the results with the same criteria: accuracy, completeness, tone, time and costs. Also count the checking time (lesson 66). A more expensive model that needs fewer corrections can end up cheaper. Repeat the test when a provider replaces a model.
Also choose based on data and reliability
Choosing a model is not just a matter of cleverness. Also look at where the model runs, which data agreements apply and whether it is available in a permitted environment (lessons 20 and 117). A local model can suit sensitive data, but requires you to manage it yourself. For fixed work processes, preferably choose a fixed model version, so that its behaviour does not change unnoticed. Note in your method card which model you used and why (lesson 100).
New practice example
From explanation to a usable result.
Choose a model per task, based on a test with your own tasks, and not just on a leaderboard.
Fictional practice situationFictional: a translation agency compares a fast model and a reasoning model on three tasks: translating a short email, translating a technical manual and checking a list of technical terms.
View a possible solution
This is how it can be done
Short email: both models good, the fast model is quicker and cheaper. Manual: the reasoning model used the technical terms more consistently, but checking is still needed. Technical terms: both models missed two terms, so a translator checks. Choice per task, with a new test as soon as a model is replaced.
Why this works
The comparison used the agency’s own tasks and the same criteria. As a result, the choice is well founded and it remains clear where checking is needed.
Make it slightly more difficult
Which three tasks from your work would you include in such a comparison, and which criterion carries the most weight for you?
Extra practice, at your own pace. You do not need to submit anything. This does not count towards an automatic assessment.
Your workplace · learning by doing
Make something. See what changes.
Checking the AI connection… You can already view the written example.
Present your assignment to Ende.
Read the answer, change one thing, and compare.
Previously held
New version
You remain in control
What do you want to improve in your next attempt?
On this device only. AI answers remain only in this page memory. This is self-review, not a grade.
Why this approach works
The comparison used the agency’s own tasks and the same criteria. As a result, the choice is well founded and it remains clear where checking is needed.
A step further
Which three tasks from your work would you include in such a comparison, and which criterion carries the most weight for you?
Try & discover · no obligation
Become the source checker yourself
Only mark statements that the source does not support. Then check your selection.
Fictional test: two models carried out three tasks. Model A was faster and just as good at rewriting. Model B made no calculation error in the figures table; model A did. In the planning, both models missed one dependency.
Select which sentences you would correct above. The answer remains hidden until you check your selection.
This trial works locally with fictitious data and does not count towards your lesson completion. It starts anew with a new lesson or reload.
A prompt to get you started
Help me set up a small model comparison for three fictional tasks: rewriting a customer email in plain language, summarising a table of sales figures and making a schedule with dependencies. For each task, draw up an assessment list with accuracy, completeness, tone, time and checking time. Explain how I can fairly compare the results of two models, without declaring an overall winner.
Apply it
Your team wants to switch to a new model because it tops a well-known leaderboard. You have not yet tested it on your own tasks.
Which approach best fits this situation?
- Switch straight away; the leaderboard proves it is better.
- At first, use the new model only for the most important clients.
- First test it on your test set with the same criteria as the current model, and check the data agreements.
With JavaScript, you can check your approach and receive immediate feedback.
Test your insight
A new model scores highest on a well-known ranking. What do you do before you use it in your work?
- Switch straight away, because it is the best model.
- Test it on your own typical tasks with the same criteria, and check the data agreements.
- Wait until colleagues complain about the old model.
Extra practice
Carry out the same three tasks with two models or modes you are allowed to use. Assess them with the same list and note for each task which model you would choose and why.
Rewriting: the fast model was done in 5 seconds and just as good; I choose it for this task. Table of figures: the fast model made a calculation error, the reasoning model did not, but it took 40 seconds; I choose the reasoning model and check the totals. Schedule: both missed one dependency, so neither without a check. Conclusion: choose per task, no overall winner.
