Sol
Use Sol for strategy, hard architecture, difficult debugging, sensitive editing, and final review when mistakes are expensive.
The plain-English answer.
Do not treat the tiers like a ranking where the biggest model must do everything. Treat them like a small team. The expensive tier should handle the few decisions that deserve it.
Terra
Use Terra for most business work, coding, research synthesis, first drafts, and routine problem solving.
Luna
Use Luna for extraction, tagging, rewriting, summaries, simple support, and batch work with clear checks.
Choose by consequence, not hype.
Ask what happens if the answer is wrong. That single question is more useful than chasing benchmark charts.
| Task | Start with | Move up when | Move down when |
|---|---|---|---|
| Business strategy | Terra | The decision is costly, public, or hard to reverse | The output is a structured checklist |
| Software work | Terra | Architecture or debugging needs deeper judgment | The task is mechanical and tests are strong |
| Content | Terra | Voice, positioning, or final editorial quality matters | You need bulk variants from an approved source |
| Data processing | Luna | Edge cases require reasoning | Stay on Luna when rules and checks are clear |
| Final approval | Sol | Already at the right tier | The work is low risk and easy to verify |
StackPilot routing rule
Plan high. Work balanced. Process cheap.
1. Plan
Use Sol for the small number of difficult choices that shape the whole project.
2. Work
Use Terra for the main draft, build, or analysis loop.
3. Process
Use Luna for repeatable substeps with explicit formats and verification.
Three common mistakes.
The model name matters less than the job definition, source quality, and review loop.
Sol for everything
You pay frontier prices for routine work that a smaller model could handle.
Luna with vague instructions
Cheap models need tighter scope, clearer examples, and better checks.
No verification
A stronger model is not permission to skip source checks or human review.
Source desk
Verify before you buy or publish.
This guide was prepared from OpenAI's July 9, 2026 GPT-5.6 release information and cross-checked against the active GPT-5.6 model available in Hermes. Product names, access, and prices can change.
Continue the cluster