Choose for the task, not the name
This lesson separates model, effort, and extended thinking, then tests whether a setting change helps. Sample outputs are teaching examples, not a benchmark of a particular model.
Model: the family doing the work. Effort: how much thinking is applied when supported. Thinking: extended thinking and its displayed summary. These are separate settings. Model availability changes with plan and account, so learn your menu rather than memorizing a roster.
Find the settings and record the test
Open a chat and click the model name near Send. The official article checked September 30, 2026 describes model selection, Effort where supported, and a Thinking or Extended toggle. Some models do not allow thinking to be turned off.
Changes apply to the next answer. An organization administrator can limit options. Record model, effort, thinking state, and prompt before comparing. Change one variable at a time and use fresh chats with the same source; earlier answers in one chat can affect later ones.
Try a routine task first
Use an available default setting:
Draft a message to employees in no more than 50 words.
Facts: the training meeting is postponed. No new date has been chosen.
Do not add a reason or date. End by saying the date will be updated after approval.
This is a draft, not for sending.
Illustrative output:
The training meeting is postponed. A new date has not been chosen
and will be shared after approval. Thank you for your understanding.
Check facts and length. If this works, more effort is not needed merely to make the wording impressive. The documentation says higher effort takes more time and tokens and can use limits faster.
Test a task with constraints
Start with the same default:
Check whether two training groups can fit today.
Each group has 24 participants. The room holds 30.
Each session takes 90 minutes, with 15 minutes of setup between them.
The room is available only from 09:00 to 12:00. Sessions cannot overlap.
Return capacity, time calculation, conclusion, and possible changes.
Do not treat a proposed change as approved or assume another room or time slot.
Illustrative correct output:
Capacity: 24 is below 30, so each group fits.
Time required: 90 + 15 + 90 = 195 minutes.
Available time: 180 minutes.
Conclusion: the plan is 15 minutes too long.
Possible changes to seek approval for: longer room availability or shorter sessions.
A common error is scheduling 09:00-10:30 and 10:30-12:00, omitting setup. Quality means keeping the constraint, not writing a long answer.
Raise effort and compare fairly
If available, try a higher effort level in a new chat with the same prompt. If testing thinking, change only that. Do not change model, prompt, and effort together and attribute the result to one setting.
Compare whether the answer preserves 24 and 30, includes setup, computes 195 versus 180, avoids silently exceeding noon, and labels changes as proposals. Note waiting time and readability too.
Check the plan against each constraint separately.
Show any conflict before proposing a replacement.
Do not remove a constraint to make the schedule look possible.
Self-checking can still fail. This calculation is small enough to verify yourself. Two answers are not proof that a setting is best for every task.
A thinking summary is not outside evidence
Thinking can display a timer and expandable summary. The documentation describes a summary, not full access to every internal step. It can help spot a missed constraint, but does not replace a source or calculation.
An incomplete summary does not prove the final answer right or wrong. Missing factual information requires a source or suitable search; higher effort does not create evidence.
Finish by choosing a setting that meets your criteria and drafting a short manager update about the scheduling conflict. Check that it does not announce an approved room extension.
Three levels of practice
Easy: identify settings
Record your model and options. Success: distinguish model name from effort level.
Intermediate: check setup time
Run the room task. Success: identify whether the 15-minute setup is counted, even in a convincing answer.
Challenging: a fair comparison
Compare two effort levels in one model using identical inputs. Success: record accuracy, time, and usefulness without claiming a universal winner.
Troubleshooting and a selection rule
No Effort option: model, plan, or organization policy may not offer it.
More thinking, same mistake: inspect source and prompt. Missing information is not an effort problem.
Too much text: specify a short deliverable and criteria. Effort and length differ.
Usage goes quickly: use defaults for routine work. Maximum effort is not required for every message.
Choose based on difficulty and constraints, then check the answer whatever the setting.