Google launches Gemini 3 Flash, makes it the default model in the Gemini app

Google contiguous released its accelerated and inexpensive Gemini 3 Flash model, based connected the Gemini 3 released past month, looking to bargain OpenAI’s thunder. The institution is besides making this the default exemplary successful the Gemini app and AI mode successful search.

The caller Flash exemplary arrives six months aft Google announced the Gemini 2.5 Flash model, offering important improvements. On the benchmark, the Gemini 3 Flash exemplary outperforms its predecessor by a important borderline and matches the show of different frontier models, similar Gemini 3 Pro and GPT 5.2, successful immoderate measures.

For instance, it scored 33.7% without instrumentality usage connected Humanity’s Last Exam benchmark, which is designed to trial expertise crossed antithetic domains. In comparison, Gemini 3 Pro scored 37.5%, Gemini 2.5 Flash scored 11%, and the recently released GPT-5.2 scored 34.5%.

On the multimodality and reasoning benchmark MMMU-Pro, the caller exemplary outscored each competitors with an 81.2% score.

Consumer rollout

Google is making Gemini 3 Flash the default exemplary successful the Gemini app globally, replacing Gemini 2.5 Flash. Users tin inactive take the Pro exemplary from the exemplary picker for mathematics and coding questions.

The institution says the caller exemplary is bully astatine identifying multimodal contented and giving you an reply based connected that. For instance, you tin upload your pickleball abbreviated video and inquire for tips; you tin effort drafting a sketch and person the exemplary conjecture what you are drawing; oregon you tin upload an audio signaling to get investigation oregon make a quiz.

The institution besides said the exemplary amended understands the intent of users’ queries and tin make much ocular answers with elements similar images and tables.

You tin besides usage the caller exemplary to create app prototypes successful the Gemini app utilizing prompts.

The Gemini 3 Pro is present disposable to everyone successful the U.S for hunt and much radical successful the U.S. tin entree the Nano Banana Pro representation exemplary successful search, arsenic well.

Enterprise and developer availability

Google noted that companies similar JetBrains, Figma, Cursor, Harvey, and Latitude are already utilizing the Gemini 3 Flash model, which is disposable done Vertex AI and Gemini Enterprise.

For developers, the institution is making the exemplary disposable successful a preview exemplary done the API and successful Antigravity, Google’s caller coding instrumentality released past month.

The institution said the Gemini 3 Pro scores 78% connected the SWE-bench verified coding benchmark, lone outperformed by GPT-5.2. It added that the exemplary is perfect for video analysis, information extraction, and ocular Q&A, and due to the fact that of its speed, it is suited for speedy and repeatable workflows.

Image Credits: Google

Model pricing is $0.50 per 1 cardinal input tokens and $3.00 per 1 cardinal output tokens. This is somewhat much costly than $0.30 per 1 cardinal input tokens and $2.50 per 1 cardinal output tokens of Gemini Flash 2.5. But Google claims that the caller exemplary outperforms the Gemini 2.5 Pro exemplary portion being 3 times faster. And, for reasoning tasks, it uses 30% less tokens connected mean than 2.5 Pro. That means overall, you mightiness prevention connected the fig of tokens for definite tasks.

Image Credits: Google

“We truly presumption flash arsenic much of your workhorse model. So if you look at, for example, adjacent the input and output prices astatine the apical of this table, Flash is conscionable a overmuch cheaper offering from an input and output terms perspective. And truthful it really allows for, for galore companies, bulk tasks,” Tulsee Doshi, Senior Director & Head of Product for Gemini Models, told TechCrunch successful a briefing

Since it released Gemini 3, Google has processed implicit 1 trillion tokens per time connected its API, amid its fierce merchandise and show warfare with OpenAI.

Earlier this month, Sam Altman reportedly sent an interior “Code Red” memo to the OpenAI squad aft ChatGPT’s postulation dipped arsenic Google’s marketplace stock successful consumers rose. Post that, OpenAI has released GPT-5.2 and a caller representation procreation model. OpenAI besides boasted astir its increasing endeavor usage and said the ChatGPT messages measurement has grown 8x since November 2024.

While Google didn’t straight code the contention with OpenAI, it said that the merchandise of caller models is challenging each companies to beryllium active.

“Just astir what’s happening crossed the manufacture is similar each of these models are continuing to beryllium awesome, situation each other, propulsion the frontier. And I deliberation what’s besides awesome is arsenic companies are releasing these models,” Doshi said.

“We’re besides introducing caller benchmarks and caller ways of evaluating these models. And truthful that’s besides encouraging us.”