Choose quality, latency and cost separately
You will inspect model/effort controls and understand fast mode and advisor as separate options. Examples are illustrations, not benchmark results.
Sources checked September 30, 2026 describe changing model rosters and organization limits. The live picker is authoritative for your account. A newer name does not prove better results for your task.
Inspect the current configuration
Use /status, /model and /effort as supported by your version. Record selected model, effort and authentication route without displaying secrets. Subscription limits and Console/provider billing differ.
Choose a supported standard configuration for this exercise. Do not use /fast as a harmless speed preference: current documentation describes higher token prices and usage credits outside subscription allowances. Its saved preference can persist across sessions.
Use one checkable problem
Plan two course groups using only these facts:
48 learners, room capacity 30, each session 45 minutes,
available window 09:00-12:00, required break 15 minutes.
Show one feasible schedule and verify capacity and total time.
Do not book, message anyone or invent staff availability.
Illustrative valid schedule:
Group A: 24 learners, 09:00-09:45.
Break: 09:45-10:00.
Group B: 24 learners, 10:00-10:45.
Each group fits 30 places. Total elapsed time is 105 minutes.
No booking or staffing approval was made.
Verify the arithmetic and boundaries independently. More effort does not turn a proposal into an approved event.
Compare without mixing variables
If two permitted standard settings are available, repeat the identical problem while changing one setting only. Record observed time, correct constraints and repairs needed. Do not invent a measured speedup.
Check the proposal against every supplied constraint separately.
Mark what is calculated, what is assumed and what remains unapproved.
Changing models or effort can rebuild prompt cache. A slower next turn may reflect that transition, not only task difficulty.
Fast mode and advisor need separate decisions
Fast mode accelerates supported Opus configurations at higher cost. It is not a new model or automatic quality improvement. Inspect current rates, billing identity and saved preference before choosing it for real work. This exercise leaves it off.
Advisor lets the main model consult another model for difficult decisions. It is additional model usage, not a free human reviewer. Current documentation describes account/model requirements and cost. Learn its role without enabling it merely to compare settings.
Graded practice
Easy: configuration
Identify current model and effort. Success: separate model, effort and billing route.
Intermediate: controlled comparison
Compare permitted standard settings. Success: identical task and observed, not invented, results.
Challenging: cost decision
Explain when latency could justify acceleration and what must be checked first. Success: no assumption that credits are free or included.
Troubleshooting
Model unavailable: organization or provider limits may apply. Do not bypass policy.
Higher effort still wrong: verify source constraints and repair the answer.
Fast mode unexpectedly on: inspect current state and saved preference before further usage.
Advisor unavailable: inspect documented requirements; do not substitute an unapproved provider.
Usage high: narrow task and context before enabling paid features.