Lesson 8 of 8 · 10 min read · Beginner → Intermediate
Cost control: the cheapest way to get the result
Routing, caching, batch, when to downgrade or cancel, and the red flags that mean you are overpaying for Gemini.
In this lesson you will
→Route each task to the cheapest Gemini model that succeeds.
→Combine caching and batch to cut a bulk job's cost by a large margin.
→Recognise the signs that a subscription or model choice is costing you more than it should.
Everything in this course points here. Gemini can be one of the cheapest ways to use a frontier-class model or one of the most expensive, and the difference is almost entirely in choices you control. This lesson is the checklist.
Figure 1.Four things drive your Gemini bill: which model, how many input tokens, how many output tokens, and how many times you run it. Fix them in that order.
1. Route by task, not by habit
The cheapest model that gets the job done is the right model. Build a simple rule: Flash Lite for classification and short rewrites, Flash for everything else by default, Pro only for tasks where Flash has demonstrably failed. On the API this is a one-line change. In the app it is the model picker. Our decision engine will suggest a tier from a task description.
Figure 2.A router sends easy requests to a cheap model and escalates only the hard ones. Even a crude rule — try Flash first, retry on Pro if the output fails a check — captures most of the saving.
2. Cut input: cache what repeats
Any prefix you send more than a handful of times should be cached. System instructions, reference documents, few-shot examples. Cached input on current Gemini models is listed at a tenth of the standard input rate in our catalogue at the time of writing. Check the storage charge and time-to-live on the official docs so you do not cache things you use once.
3. Cut output: ask for less
Output tokens cost several times input on every Gemini model. "Bullets only", "return JSON", "no preamble", and a maximum output token setting on the API all reduce the expensive side. If a task needs a long answer, that is fine — but do not pay for a long answer to a short question.
4. Cut volume: batch and dedupe
If the answer can wait, use batch mode; the batch rows in our catalogue are priced at half the standard rate on the models we track. Before that, check whether you are sending duplicates: the same document summarised twice, the same query from two systems. Deduplication is free.
Figure 3.The same job — say, summarising two thousand documents — costs $32 on Pro and about $1.40 on Flash Lite. Routing is worth more than any other single optimisation. Add caching on the Pro rows and the gap narrows but does not close.
5. Downgrade or cancel on evidence
On Google AI Ultra and not using video generation or hitting Pro limits weekly: drop to Pro. That is a saving of more than $229 a month at catalogue prices.
On Google AI Pro and not using Gemini inside Docs, Gmail or Sheets, and not hitting free-tier limits: cancel and use the free app. Spend the odd dollar on the API for big jobs.
On the legacy Google One AI Premium plan: check whether AI Pro at $19.99 gives you the same or more for a similar price and switch if so.
Paying for two chat subscriptions (Gemini plus ChatGPT or Claude): pick one for a month and see what you miss. See ChatGPT vs Gemini and Claude vs Gemini.
A team spends $400 a month on Gemini 3.1 Pro Preview summarising internal documents overnight. Which change saves the most?
Lesson FAQ
▸How do I see what I spend on the Gemini API?
Google AI Studio shows usage, and Google Cloud billing shows cost once billing is enabled. Log usageMetadata from each response as well, so you can attribute spend to tasks and models.
▸Is a Gemma model a realistic replacement for Flash?
For plain text tasks at high volume, often yes, and it can be run privately. For multimodal input, very long context or the hardest reasoning, no. Test on a sample before switching.
▸Where do I go next?
Compare Gemini against the alternatives in [Claude 101](/tutorials/claude-101) and [ChatGPT 101](/tutorials/chatgpt-101), or run your numbers through the [calculator](/calculator). The [AI pricing tracker](/insights/ai-pricing-tracker) shows how Gemini prices have moved.
Finished reading?
Mark it done to track your progress through the course.
If our calculators helped you cut down on hidden AI wallet leaks, consider buying us a coffee. A tiny fraction of your savings keeps our pricing indexes updated daily.
Not sure which AI is cheapest for your use case? Find out in 30 seconds — no signup required.