Keep a weekend project on free Gemini
Scope a weekend side project to free Gemini for drafts, brainstorming, and first-pass code, and upgrade only when recurring reasoning blocks delivery.
A weekend build should not start with a subscription. Put drafts, brainstorming, and first-pass code on the free Gemini model, and pay only when recurring advanced reasoning blocks the deliverable.
Google's Gemini model-access changes for users without an AI subscription begin on October 9, 2026. That date matters because the free app becomes more predictable. After that date, free-account users in the Gemini app are restricted to Gemini 3.5 Flash-Lite, and the 3.6 Flash and 3.1 Pro options are no longer available to them.
For a side project, that is a useful constraint. The free model is fast enough for the work that keeps a weekend moving: naming features, sketching copy, and producing code you can run, break, and improve. Gemini 3.5 Flash-Lite is positioned as the fastest model in the 3.5 series, and Artificial Analysis measured its output speed at 350 tokens per second.
Auto model selection will be the default for Gemini users without a Google AI plan. Under that setting, most prompts will be routed to Flash-Lite. Turning off smart model selection makes every response come from Flash-Lite. The free experience becomes one predictable path, which is easier to scope.
The free model carries the weekend work
Most weekend projects die from scope, not model quality. A landing page, a small tool, a newsletter draft, or a prototype can all get a first pass on the free model. Paid models earn their place when the work needs sustained reasoning: a complex refactor, a multi-step debugging session, or a plan that must hold together across many files.
The budget test is simple. Keep one-off tasks on the free model. Let recurring tasks that block the deliverable justify paying. Let occasional tasks wait until the next weekend.
The checklist is the budget test
Keep this list next to your project notes. Each item is quick to check.
- Open the model picker and confirm the free model is selected. A picker with only one model means you are already using the free model.
- Put drafts, brainstorming, and first-pass code on the free model. Use it for work that needs speed and iteration, not final proofing.
- Upgrade only for recurring advanced reasoning that blocks the deliverable. A paid model needed every session, with the project unable to move without it, makes the cost a project cost rather than a hobby expense.
The first item catches an interface that can hide the constraint, the second matches the free model's strength for early work, and the third keeps a side project from becoming a subscription experiment.
The price becomes real when reasoning blocks delivery
Paid tiers change what you can keep. The $4.99-per-month AI Plus plan will lose access to Gemini 3.1 Pro but keep access to Gemini 3.6 Flash. The $19.99-per-month AI Pro tier will gain access to the Deep Think option that had been available only on the $99.99 and $199.99 plans. Deep Think is a reasoning feature that had previously been limited to higher tiers. Maintaining the pre-change free-account experience in the Gemini app requires at least $19.99 per month for Google AI Pro.
For a weekend project, the $4.99 tier is a middle path when you need Flash but not Pro. The $19.99 tier is the step to take when the work needs sustained reasoning, and the monthly cost is easier to justify than a stalled prototype.
Usage limits also shape the decision. Gemini's usage allowance is compute-based, resets every 5 hours, and is capped by a weekly limit. Short building bursts fit a weekend. Long sessions can hit the weekly cap and slow the project.
An API route is a different budget. Through Google's API, Gemini 3.5 Flash-Lite costs $0.30 per million input tokens and $2.50 per million output tokens. For a small app with light traffic, that can stay small. For a public tool with steady users, it can become a real line item before the project ever makes money.