You watched the video — here's the whole system. OpenAI just split its newest model into three price tiers, and the gap is 5× per token inside the same family. The fix is a ten-second habit: match the tier to the task. By the end of this page you'll have the cheat sheet and a copy-paste prompt that picks the right tier for you.
Why you're overpaying
Last week OpenAI split its newest model into three separate price tiers — Sol, Terra, and Luna — instead of selling one do-everything model.
The flagship, Sol, runs $5–$30 per million tokens. The cheap one, Luna, is $1 in and $6 out. Same company, same model family — five times the price.
That split tells you something most people miss: you don't need the best model for most of what you do. And you're probably paying flagship prices for work a cheaper brain does just as well.
The full step-by-step
1. Notice the tier you're about to use. Most people default to the most powerful model for everything — that's the leak.
2. Ask one question before the task: is this boring work (summarize, reformat, draft an email) or genuinely hard thinking (debugging, strategy, multi-step reasoning)?
3. Boring → switch to the cheap/fast tier. Hard → keep the flagship. In between → the everyday tier.
4. Only bump up when an answer is visibly worse. If the cheap tier nailed it, you just saved the difference.
The cheat sheet
| Tier | Use it for | Claude | Gemini | OpenAI (GPT-5.6, rolling out) |
|---|---|---|---|---|
| Cheap / fast | summaries, emails, reformatting, quick lookups, extraction | Haiku | Flash | Luna ($1 in / $6 out per 1M) |
| Everyday | drafting, normal chat, decent reasoning, day-to-day work | Sonnet | Flash (thinking) | Terra |
| Flagship | debugging, strategy, multi-step reasoning, long hard problems | Opus | Pro | Sol ($5–$30 per 1M) |
Those OpenAI tiers are still rolling out — but the tools you already pay for are tiered today, so the habit works right now.
The exact prompt
Not sure which tier a task needs? Paste this before any task and let the model decide:
Rate this task from 1–10 for complexity.
Then tell me the cheapest AI tier (cheap / everyday / flagship)
that can still do it well — and why.
Task: <paste your task>
If it says 1–4 → use the cheap tier. 5–7 → everyday. 8–10 → flagship.
Tools & links
- Claude — Haiku / Sonnet / Opus tier picker lives in the model dropdown: claude.ai
- Gemini — Flash vs Pro, same move: gemini.google.com
- ChatGPT — the Sol / Terra / Luna tiers are rolling out; watch your model picker: chatgpt.com
- The video — the 60-second version of this page: watch it on YouTube
What most people get wrong
They treat the model picker as a quality dial and leave it maxed out. It's not a quality dial — it's a price dial with a quality floor. For most tasks the cheap tier clears the floor, and the flagship just costs 5× more to produce the same email.
One step to take today
Set your everyday assistant to the cheap/fast tier right now, and only bump up when an answer is visibly worse. That single default change is where almost all the savings live.
*Get genuinely better — daily.*