What Is the Free Model of Google Gemini?

What Is the Free Model of Google Gemini?

Gemini 3.5 Flash-Lite. That answer changed very recently, and if you read this question anywhere written before October 2026 you will have been told something different.

On 9 October 2026 Google narrowed what the free Gemini plan can reach. Users without a paid Google AI subscription lost access to the Flash and Pro models they had been using, leaving Flash-Lite as the model behind the free tier.

This guide covers what Flash-Lite actually is, what the change took away, how the free plan compares with the paid tiers now, and how to tell which model answered you — because the interface does not always make that obvious.

The short answer

The free plan runs on Gemini Flash-Lite, currently version 3.5. Before 9 October 2026 a free user could also reach Flash and, in some flows, Pro. Those two are now reserved for paying subscribers.

Google's own model-availability table is the authority on this, and it is worth bookmarking rather than trusting any article including this one, because the table has been revised twice in 2026 already. The short version of that table reads:

• No plan: Flash-Lite
• Google AI Plus: Flash-Lite and Flash
• Google AI Pro: Flash-Lite, Flash and Pro, plus the Deep Think option
• Google AI Ultra: Flash-Lite, Flash and Pro, with the highest limits

One wrinkle is worth flagging honestly. The free plan still defaults to Auto, Google's smart model selection, and Google describes Auto generically as sending "each prompt to the Gemini model that suits it best." That description is written across all plans, so what Auto can actually reach is bounded by your plan's row in the table rather than by the sentence describing Auto. On a free account, that bound is Flash-Lite.

Curious whether AI assistants are recommending your brand? Get a free SEO audit and find out how your site looks to them.

What is Flash-Lite, and is it any good?

Flash-Lite is the smallest and fastest model in the Gemini line. It is built for high-volume, low-latency work: short answers, summarising, extraction, classification, rewriting. Google positions it as the cost-efficient tier, which on the consumer side translates to the one it can afford to give away.

For a large share of everyday use that is genuinely fine. Asking for a definition, tidying up an email, summarising a page you have pasted in, converting a list into prose — Flash-Lite handles all of that without any obvious shortfall.

Where you will notice the difference

The gap opens up on anything that needs sustained reasoning rather than fast recall.

• Multi-step problems where an early mistake compounds, such as longer arithmetic or logic chains
• Code that has to hold several files or constraints in mind at once
• Long documents, where the smaller model loses the thread sooner
• Nuanced writing, where output tends towards the generic
• Anything you would describe as "think about this carefully"

The practical tell is confident wrongness. A smaller model rarely says it is unsure; it produces a fluent answer with a flaw in the middle. If you are using the free plan for work that matters, the habit worth building is verifying the steps rather than the conclusion.

One more caveat on judging it: much of what people read as model quality is prompt quality. A vague request gets a generic answer from any model, and Flash-Lite punishes vagueness harder than a larger model does because it has less headroom to infer what you meant.

Speed is a real advantage, not a consolation

It is worth saying that Flash-Lite being fast is not merely a euphemism. For iterative work — rewriting a paragraph eight times, testing phrasings, batch-processing short inputs — a model that answers in under two seconds is more useful than a slower, smarter one. The free tier is not a crippled version of the product so much as a different point on the trade-off.

What exactly changed on 9 October 2026?

Three things, and they affected different users differently.

Free users lost Flash and Pro. That is the headline change and the one that prompted most of the complaints: workflows people had built on the free tier got quietly less capable.

Google AI Plus subscribers, on the $4.99 tier, also lost something. Pro is no longer part of that plan, leaving Flash-Lite and Flash. Google said affected subscribers would receive an email explaining when the change takes effect for them.

The pattern is worth naming because it will probably repeat. Capability moves up the tiers over time while the price points stay put, so the plan you chose a year ago is not the plan you have today even if your bill has not changed.

AI Pro and AI Ultra subscribers lost nothing, and AI Pro gained the Deep Think option. The change was a re-tiering that pushed capability upward, not a cut across the board.

If you publish technical content, AI assistants are already deciding whether to cite it. Run a free audit to see how yours is positioned.

How do the free plan's limits work?

Separately from which model you get, there is how much of it you get — and Gemini's limits are not counted in messages.

Google moved to compute-based limits, which means a prompt's cost depends on its complexity, the features it uses and how long the conversation has become. Two prompts that look similar can consume very different amounts of your allowance.

Google describes the tiers relative to each other rather than in absolute numbers:

• No plan: standard limits
• AI Plus: 2x higher than standard limits
• AI Pro: 4x higher than standard limits
• AI Ultra: 5x or 20x higher than AI Pro, depending on the subscription

Limits refresh every five hours until you reach a weekly ceiling. Media generation, Deep Research and extended thinking all draw more heavily on the allowance, and the effort levels Google has been rolling out — low, medium and high — consume more of it at the higher settings.

A practical consequence: long conversations get expensive. Because chat length feeds into the compute cost, the cheapest way to stay inside the free allowance is to start a fresh chat for each new task rather than carrying one thread through the day.

Google does not publish a numeric prompt cap for any plan, which is deliberate and genuinely inconvenient. You find your limit by hitting it.

How do you tell which model answered you?

This matters more than it sounds, because the most common complaint about any AI assistant — "it got worse this week" — is usually a model switch rather than a model regression.

Check the model picker above or beside the prompt box; it shows what is selected and, on Auto, what the system chose. If the picker offers you Flash or Pro and you are not paying, you are most likely looking at a stale client, so refresh the page or update the app.

There is a third check that costs nothing: ask it. Gemini will usually name the model it is running, and while a model's self-report is not authoritative, a mismatch between what it says and what the picker shows is a useful prompt to look closer.

The second check is behavioural. If answers have become shorter, flatter and faster, and the reasoning steps you used to see have disappeared, you have been moved to a smaller model. On a paid plan that usually means you have hit a cap and been downgraded for the rest of the window.

When is the free plan genuinely enough?

For most individual, non-professional use, it is. If Gemini is where you go to draft messages, summarise articles, get explanations and ask quick factual questions, Flash-Lite will not hold you back, and the Google Search grounding that makes Gemini useful for current information is available without paying.

It is also enough for evaluation. A fortnight on the free plan tells you whether Gemini fits your habits, which is more informative than any comparison. The limits you hit are the signal about which plan you need.

Not sure which of your pages AI tools actually surface? Start with a free SEO audit and work from data rather than guesswork.

Where it stops being enough is fairly predictable: sustained coding, long-document work, research tasks with several steps, or any use where a quiet error is expensive. If you are hitting limits mid-task several times a week, that is the upgrade signal rather than any feature on a comparison page.

Free model tiers shift without warning. Your search visibility should not. Get a free SEO audit and see where you stand.

What are the limitations worth knowing first?

The model tier can change under you. It has changed twice in 2026, and nothing commits Google to keeping Flash-Lite on the free plan either. Build a habit on a free tier if you like; do not build a business process on one.

Compute-based limits are hard to plan around. Without a published number you cannot budget your usage, and the heavier features burn through the allowance faster than ordinary chat.

Feature access does not track model access neatly. Some capabilities are gated by plan independently of the model, so "which model do I get" and "what can I do" are two different questions with two different answers.

Context length is the quiet constraint. The smaller model's practical working memory is shorter, so very long conversations degrade — and the degradation is gradual rather than announced.

And the version numbers move. Flash-Lite 3.5 is the current free model at the time of writing; the useful thing to remember is the tier name, not the number, because the number will change while the tier stays put.

Gemini's free plan versus the other free tiers

Against ChatGPT's free tier, the two are closer than either company's marketing suggests. Both give a capable small model, both include web grounding, both cap usage. Gemini's edge is Google Search integration and the speed with which Google pushes features downward to free users; ChatGPT's is a slightly stronger baseline model and a larger ecosystem of custom tools.

Against Microsoft Copilot's free tier, the comparison is less about the model and more about where you work. Copilot's advantage is that it sits inside Office documents you already have open; Gemini's is the same argument applied to Google Workspace. On raw free-tier capability they are close enough that the surrounding software decides it.

Against Claude's free tier, Gemini wins on breadth and search grounding, and Claude tends to win on long-form writing quality and on holding a long document together. If your work is mostly reading and drafting prose, that trade is worth testing directly.

Against paying $4.99 for AI Plus, the honest answer is that the gap narrowed on 9 October and then widened again. AI Plus now buys Flash and double the limits but no longer Pro, which makes it a limits upgrade more than a capability upgrade. If the reason you were considering it was Pro-level reasoning, AI Pro is the tier that actually delivers that.

Final Thoughts

The free model of Google Gemini is Flash-Lite, and as of 9 October 2026 that is the whole of what a free account can reach. It is a fast, capable small model that covers ordinary daily use comfortably and runs out of depth on anything that needs sustained reasoning.

The more durable lesson is that free tiers in this market are a moving target. Three models became one in a single announcement, with no change to the product name and no prompt to the user. Check the model picker occasionally rather than assuming what you signed up for is what you are still using.

If you are weighing Gemini against the obvious alternative rather than against its own paid tiers, our Gemini versus ChatGPT comparison goes through it use case by use case and reaches a verdict for each one.

And if you publish anything you want these assistants to cite, the model powering their free tier matters less than whether your pages are structured for them to find in the first place.