Free Gemini Now Runs on 'Auto', and Google Picks Its Lightest Model

Published: October 21, 2026 Impact: High. Affects every Gemini user without an AI subscription.

On October 9, 2026, Google changed the free Gemini app. Its help page says: "Most prompts will go to Flash-Lite, our fastest model. Prompts that need deeper reasoning may go to Flash or Pro." Free users no longer pick a model. The app runs on Auto and does not show which model answered.

Free Gemini now defaults to Google's lightest model, and Google does not tell you when you get a better one.

AI researcher Andrej Karpathy, writing on X on Oct 11, put the free tier this way: "Around 1-2B (~20%) are casual and infrequent users of free-tier ChatGPT-like products. This group sees derpy chatbots". The largest free tier just moved a step in that direction.

What Happened

Google's help page, Changes to Gemini model access and limits, states: "These changes will start to take effect for users without an AI subscription on October 9th." Without a plan, you get Auto across Flash-Lite, Flash and Pro. AI Plus gets Flash-Lite and Flash, with no Pro. AI Pro and Ultra get all three. Google adds: "If smart model selection is turned off, chats for users without an AI plan will use Flash-Lite for all responses."

The usage limits did not change on October 9. The compute-based limits, a refresh every 5 hours plus a weekly cap, date from May 17, 2026. October 9 changed which model answers, not how much you can ask.

The record so far:

• May 28: Google said it would give users a more detailed breakdown of their usage, as 9to5Google reported.
• July 21: Google called 3.6 Flash "Our workhorse model that delivers better coding, knowledge work, and multimodal performance" and made it available "For everyone via the Gemini app."
• July 27: OneHuman's logged-in check found only a two-bar meter, with no breakdown by prompt, feature or model (earlier report).
• October 9: the model itself became invisible.

OneHuman's owner checked a free account on October 21:

"I chose free Gemini on my laptop and phone deliberately. It was powerful, it was really free, and I rarely hit a limit. On October 9 Google started choosing the model for me. The app just says 'Auto'. It never tells me which model answered, and the better ones now sit behind a padlock and an Upgrade button." (the owner)

The model menu offers one usable option, Auto. 3.1 Pro and 4 Argon both show a padlock.

Clicking a locked model opens an Upgrade pop-up.

What This Means

Google's page says free prompts "may go to Flash or Pro". The free app shows Pro as locked and never says when Auto used it. I am not guessing why. The gap is there and it is documented.

Fairness first. By Google's own numbers, 3.5 Flash-Lite beats the older Gemini 3 Flash on SWE-Bench Pro (54.2% against 49.6%) and on OSWorld-Verified (74.0% against 65.1%). Flash-Lite is not a bad model. Free users now mostly get Google's light model by default.

The trade-off is smarter against faster. On Artificial Analysis, Gemini 3.6 Flash (High) scores 34 on the Intelligence Index and Flash-Lite scores 22. Flash-Lite outputs 454 tokens a second against 214, about twice as fast.

Quick lookups and rewrites barely change and may feel faster. Multi-step questions, charts, photos and planning are where a lighter default shows.

Long documents are the hard limit. The free context window is 32k tokens. AI Plus gets 128k, and AI Pro and Ultra get 1 million. That window is not new on October 9. Google's own example of a large file prompt is "Analyze this 100-page PDF financial report." Artificial Analysis equates a million tokens to about 1,500 A4 pages, roughly 670 tokens a page. A 100-page report is therefore about 67k tokens, roughly twice the free plan's 32k window, so it can exceed what free Gemini reads at once.

Consumer Protection Q&A

Q: Did Google cut the free limits on October 9? A: No. The limits date from May 17. October 9 changed which model answers.

Q: Can I choose a model on a free account? A: No. Auto is the only unlocked option. You can turn smart model selection off, but then every answer comes from Flash-Lite, in Google's words.

Q: Does the app tell me which model answered? A: In the owner's free account, no model name appeared on the answer.

Q: Did Google notify free users? A: Google's page says AI Plus subscribers should receive an email. It says nothing about notifying free users, and I make no claim either way.

Q: Is free Gemini now bad? A: No. The default is the lighter model, and the app does not say when you get more.

What Happens Next

• 30 days: watch whether Google adds a model label to Auto answers or a per-model usage breakdown, promised in May and absent in July.
• 90 days: watch whether the free 32k window or the Plus 128k window changes.
• 6-12 months: watch whether "Auto" becomes the default on paid plans too. Google's table already limits Plus to two models.

What to Do Next

• If you need consistency, turn Auto off. Every answer is then Flash-Lite.
• If you work with long documents, the free window is 32k. Plus at $4.99 a month gets 128k and Flash but no Pro. Pro gets 1 million.
• Before paying anywhere, compare the free plans of other tools on the Compare page.

Bottom Line

What's real: Google's own page says most free prompts go to Flash-Lite, and the app does not show which model answered. What to watch: a model label, or the per-model usage breakdown promised in May. What's next: The best AI deal isn't always paid, and the best free deal can change overnight. OneHuman tracks both, dated and sourced. Join free.

Sources

• Google Help, Changes to Gemini model access and limits, checked October 21, 2026.
• Google Help, Gemini Apps limits & feature availability, checked October 21, 2026: context windows and usage multiples.
• Google, Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber, July 21, 2026.
• Artificial Analysis, Gemini 3.6 Flash (High) vs Gemini 3.5 Flash-Lite, checked October 21, 2026, Intelligence Index v4.3.2.
• Andrej Karpathy, X, October 11, 2026, via OneHuman's repost.
• OneHuman, free Gemini account screenshots, October 21, 2026.
• OneHuman earlier reports: Gemini usage transparency promise, three quiet moves in 60 days, pricing expiry and usage cap disclosure.

Verified by OneHuman · October 21, 2026

Independent. AI-assisted. Human-verified. No ads. No affiliates. No investors.