Which Gemini Model Should You Use?
Start with Flash for everyday work if your plan lets you choose, and move up to Pro or Argon only when you can see the answer falling short. On a free account you can't really choose: you get auto selection or Flash-Lite. Thinking level is the second dial, and it defaults to Low.
Here is a task I plan to give you: [describe the task and paste one real example] I can choose Flash, Pro or Argon, with Low, Medium or High thinking. Recommend the lightest combination likely to do this well. Tell me what I'd probably lose by going one step lower, and give me one check I can run to compare the answers.
- Flash-Litequick answers, lightweight tasks
- Flashstandard chat and tasks
- Procomplex prompts, advanced coding
- Argoncomplex workflows, top performance
Try it now
- 1Open a chat and click the model name at the bottom of the chat box. Models with a lock icon need a plan upgrade.
- 2Choose a model, and optionally Low, Medium or High. Your choice stays in that conversation.
- 3Run a real task, then open a new chat with a stronger setting and paste the same prompt.
- 4Under each response, click More, then Show thinking steps, and scroll to the bottom to see which model answered. If you can't point to a difference in the answers, keep the lighter one.
Use it when
Whenever a task is hard and you'll act on the result, such as comparing contracts, tracing a bug across files or planning a multi-step project. It's also worth a look when you've repeated a task several times and the answers keep missing the same thing.
Skip it when
Short everyday questions. Leave the default and move on. If you're on the free plan, the useful moves are a better prompt and a higher thinking level, since the model picker needs a Google AI plan.
Why it works
Google's models page lists Flash-Lite, Flash, Pro and Argon. With no plan you get Auto or Flash-Lite. Google AI Pro and Ultra get all four in chat. On Plus, the sources differ: the models page lists Flash-Lite, Flash and Pro, while the plan page names only Flash, so check your own picker. Every plan gets Low, Medium and High thinking. Google says limits are based on compute and depend on the model and features used, so stronger settings use your allowance faster. Argon reached the app on October 13, 2026, for Pro and Ultra subscribers.
Example
Rule of thumb, not a Google chart. Say you have a three-paragraph email to rewrite and a 60-page supplier contract to check against last year's. The email is Flash on Low. The contract is worth running on Pro with High thinking, and on Argon if you have it, because a missed clause costs more than a slower answer. Attach both contract versions and run the same prompt on each setting before deciding which to use for the rest of the year's renewals.
Going further
Check the toggle at Settings > Usage limits. It controls auto model selection, and Google says most prompts go to Flash-Lite when it's on. Watch for one more trap on Ultra: Deep Think uses a 192 thousand token context window, not the 1 million of other top settings, so it suits hard reasoning more than big files. Google also says Argon is replacing Deep Think.
Common mistakes
- Believing the answer came from the model you picked after you've hit a limit. Google says the chat continues on Flash-Lite, so a sudden drop in quality can be that.
- Using High thinking for everything, then running out of allowance. Save it for prompts that have several steps.
- Comparing models with different prompts. Change one thing at a time, or the test tells you nothing.
- Following tutorials with different labels. Google's Deep Research page still says Thinking and Pro, while its models page lists four other names. Go by your own picker.
Sources
- About Gemini models, Gemini Apps Help
- Gemini subscriptions
- Gemini Apps limits and feature availability
- Gemini Apps release notes
- Upload and analyze files, Gemini Apps Help
- Use Deep Think in Gemini Apps
- Deep Research, Gemini Apps Help
Checked 6 November 2026. Products change often, so confirm details with the vendor.