What is Gemini 3 Flash, and why is it now the default model in Gemini?

Yesterday, Wednesday, December 17, 2025, Google released Gemini 3 Flash, the fast member of its Gemini 3 family. Last month Google introduced Gemini 3 Pro and a Deep Think mode, its most capable models. Gemini 3 Flash aims to deliver much of that intelligence at a lower price and with shorter waits, and Google is making it the model most people will meet first.

Starting now, it replaces Gemini 2.5 Flash as the default in the Gemini app worldwide, and it is rolling out as the default model behind AI Mode in Google Search, the conversational search feature. That means free users get the Gemini 3 generation without paying.

What Google released

Google sorts its models into tiers. Pro models are the largest and most thorough. Flash models are built for speed and cost, the kind of model companies run millions of times a day. Tulsee Doshi, who leads product for Gemini models, told TechCrunch that Google sees Flash as the workhorse, suited to tasks done in bulk.

Gemini 3 Flash is available in several places at once:

• For everyone: at no cost in the Gemini app and in Search's AI Mode. In the app, Pro remains one tap away in the model menu for math and coding questions.
• For developers: a preview through the Gemini API in Google AI Studio, plus the Gemini command-line tool, Android Studio and Google Antigravity, the agent-based coding tool Google released last month.
• For companies: Vertex AI and Gemini Enterprise, Google's business platforms. Google names JetBrains, Figma and Bridgewater Associates among early users.

For developers, the price is $0.50 per million input tokens and $3 per million output tokens. A token is a small chunk of text, roughly part of a word. That is a little more than Gemini 2.5 Flash, which cost $0.30 and $2.50, but Google argues the new model often needs fewer tokens to finish a job.

How fast models think

Speed in AI is mostly about latency, the wait between your question and the start of the answer. A smaller, efficient model computes each token faster, which matters when a reply must appear instantly in an app or a search page.

Gemini 3 Flash is also a thinking model: it can reason through a problem before answering. Google says it adjusts how much it thinks to the task, reasoning longer on a hard question and briefly on an easy one. On typical requests, Google measured it using about 30% fewer tokens than Gemini 2.5 Pro while performing better. Fewer tokens means a faster answer and a lower bill.

A useful way to picture this is a chart with quality on one axis and cost or speed on the other. Every model sits somewhere on it, and the best ones form an edge where you can't get more quality without paying more. Google says Gemini 3 Flash pushes that edge outward, giving better results at a given cost.

The numbers

These are Google's reported scores, with comparisons from TechCrunch:

Google also reports that it beats Gemini 2.5 Pro, its flagship until last month, while running about three times faster according to the independent benchmarking firm Artificial Analysis. On Humanity's Last Exam, the previous Flash model scored 11%, so the jump within one generation is large.

A worked example

Gemini 3 Flash is multimodal: it can take in video, images and audio as well as text. Google's own examples include analyzing a short video of a golf swing, guessing a drawing while you sketch it, and turning a recorded lecture into a quiz. Here is a request in that spirit:

I've uploaded a 2-minute recording of my lecture on photosynthesis.
1. List the three main ideas in plain language.
2. Point out anything I explained unclearly or skipped.
3. Write a five-question quiz with answers, mixing easy and hard.

For quick tasks like this, a fast model shines: you get a useful draft in seconds, and you can refine it in a few rounds without long waits.

Why it matters

Default models matter more than flagships because they are what most people actually use. By putting a Gemini 3 model behind both the free Gemini app and AI Mode, Google is bringing its newest generation to a very large audience at once. Google also says its API has processed more than a trillion tokens a day since Gemini 3 launched.

The release lands in a fast-moving contest. TechCrunch notes that in the past week OpenAI shipped both GPT-5.2 and an updated image generator after an internal "code red" memo about Google's progress. Gemini 3 Flash answers with a different argument: near-flagship quality at a price that makes heavy everyday use affordable, both for consumers and for companies building products.

For developers, the lesson of the past year continues: smaller models keep closing the gap with larger ones. A model that matches last spring's flagship at a fraction of the cost changes which applications make economic sense.

When should you choose Gemini 3 Flash over Gemini 3 Pro?

Use Gemini 3 Flash for most daily work: questions, summaries, writing help, quick analysis of a photo or video, and planning. It is the default for a reason, and its speed makes back-and-forth refinement pleasant. For developers, it suits apps that need fast replies and tasks run at high volume, such as sorting documents or extracting data.

Switch to Gemini 3 Pro when the problem is hard and mistakes are costly: difficult math, complex coding across many files, or deep analysis where you want the most careful reasoning available. Google itself points users to Pro in the model menu for math and coding.

A practical approach is to start with Flash, and move to Pro only when an answer seems shallow or wrong. That way you spend time and money where the extra depth actually helps.