Draft on a cheaper model, review on a stronger one

The scene

You have a lot of drafting, sorting, translation, or classification to do, but you want a more careful pass before you ship writing or code.

Tools involved

What they did

1. Split the job into a draft/extract/classify pass and a review/finalize pass. Do not run the most expensive model over the whole task.
2. For the first pass, pick a free-tier or lower-priced API. Google’s AI Studio pricing page lists Gemini 3.5 Flash-Lite at $0.30 per million input tokens and $2.50 per million output tokens (including thinking), with a free tier. DeepSeek’s docs list deepseek-flash off-peak cache-miss input at $0.15 per million tokens and output at $0.60 (peak rates are double).
3. For the second pass, switch to a stronger model. Anthropic’s official pricing lists Claude Sonnet 5.5 at $2 input / $10 output per million tokens, and Claude Opus 5.5 at $4 / $20. OpenAI’s official pricing lists standard gpt-5.4-mini at $0.75 / $4.50, and short-context gpt-5.4 at $2.50 / $15.00.
4. Hand the first pass to the second as source material. Ask it only to fix errors, flag uncertainty, leave sound text alone, and invent no figures.
5. A person should still review anything that will be sent, submitted, or written into a formal file. A model stack can cut cost; it cannot sign off for you.

Copyable prompt

I will give you a draft produced by another model. Review it against the files or notes I also attach. Fix only errors and unclear claims, mark anything you cannot verify, and do not invent numbers. Keep the original structure unless a change is necessary. Return: (1) the revised draft, (2) a short list of changes, (3) a list of items that still need a human decision.

Rewritten here for you to try; not a verbatim excerpt from elsewhere. What it asks：Give a stronger model the cheaper draft plus the original materials. It should fix problems, flag what it cannot check, and list decisions that still need a person.

Things to watch

These are official standard API list prices per million tokens, before tax, and may not match your final bill. Cache hits, batch tiers, context length, region, and free-tier caps all change what you pay. DeepSeek’s peak hours are 01:00–04:00 and 06:00–10:00 UTC, Monday–Friday, excluding Chinese public holidays. Gemini 3.8 Flash paid rates are listed as $0.75 input / $3.75 output through 31 Dec 2026, then $1.50 / $7.50 from 1 Jan 2027. Official pages change; use the date on this article.