Gemini Swapped ‘Extended’ for Low, Medium and High Thinking. Here’s When to Use Each

Open the Gemini app on your Android phone this weekend and the model picker may look different. Google is rolling out a new set of thinking levels for subscribers, replacing the old “Extended” toggle with three settings: Low, Medium and High. At the same time, free users lost the picker almost entirely. Their only option now is “Auto.”

It’s a small menu change with real consequences for how fast, and how carefully, Gemini answers you. Here’s what changed and how to use it.

What changed on October 9

9to5Google spotted both changes Thursday evening. For subscribers, the new thinking levels are rolling out now and are meant to line up with the options Google already offers developers in AI Studio and its Antigravity coding tool. Google describes them simply:

• Low: quick and efficient
• Medium: balanced depth
• High: extra thorough

Free users can no longer switch models manually. The “Auto” option, described as using “Gemini 3 models,” routes each prompt to the model Google thinks fits best. Most prompts go to Flash-Lite, Google’s fastest model, while prompts that need deeper reasoning may get bumped to Flash or Pro. Turn off smart model selection and you get Flash-Lite for every response. NewsBytes reports most free accounts had the change by Friday.

One more change is coming: Google AI Plus subscribers will soon be limited to Flash-Lite and 3.8 Flash.

What thinking levels actually control

A thinking level sets how much reasoning the model does before it writes its answer. More thinking usually means better results on hard problems, and slower replies. Google’s developer documentation is the clearest guide to the trade-off. It recommends minimal or low thinking “for fact retrieval or classification,” the default level for moderate tasks like comparing concepts or creative work, and maximum thinking “for advanced coding, math, or multi-step planning.”

The same docs show that Google tunes the default differently for each model, which hints at how it sees them:

That’s the API, not the consumer app, and Google hasn’t said exactly how the app’s three levels map onto it. But the pattern is a useful rule of thumb: Google treats Flash-Lite as a no-frills answer engine and Pro as the deep thinker.

When to use Low, Medium or High

Low

Use it for anything you’d have typed into a search bar: a quick definition, a unit conversion, rewording a text before you send it, sorting a list. Speed matters more than depth here, and extra reasoning mostly adds waiting.

Medium

This is the sensible everyday setting. Comparing two phone plans, drafting an email that needs some tact, summarizing a long article or brainstorming gift ideas all benefit from some reasoning without needing the full treatment.

High

Save it for work that falls apart if a step goes wrong: debugging code, checking math, building a multi-day trip itinerary with real constraints, or untangling a spreadsheet formula. Expect to wait longer. That’s the point.

The free tier got the short end

Thinking levels are a genuine improvement for paying users. “Extended” was a blunt on-off switch, and three levels give you real control over the speed-versus-quality trade-off. The free tier’s version of control, by contrast, is a toggle that can only make things worse: switch off smart selection and you’re on Flash-Lite for everything.

Google’s bet is that Auto routing is good enough that most free users won’t notice, and for quick questions it probably is. But taking the model picker away while handing subscribers finer controls draws a clear line. If you rely on Gemini for coding, math or long planning sessions, the free tier is no longer built for you. If you use it like a smarter search box, Auto will do fine, and you can leave the settings alone.

Sources: 9to5Google, NewsBytes, Google AI for Developers. Photo by Shutter Speed on Unsplash.