In an attempt to overtake OpenAI, Google today unveiled the Gemini 3 Flash model, which is quick and inexpensive and is based on the Gemini 3 that was launched last month. Additionally, the business is making this the default model in search AI mode and the Gemini app.
Six months after Google unveiled the Gemini 2.5 Flash variant, the new model offers notable enhancements. The Gemini 3 Flash model scores far better on the test than its predecessor and, in many cases, is comparable to other frontier models, such as the Gemini 3 Pro and GPT 5.2.
For example, it received a score of 33.7% on Humanity’s Last Exam standard, which is intended to assess proficiency in several subjects, without the use of any tools. By contrast, Gemini 2.5 Flash scored 11%, Gemini 3 Pro scored 37.5%, and the recently launched GPT-5.2 scored 34.5%.
With an 81.2% score, the new model beat all rivals on the multimodality and reasoning test MMMU-Pro.
Rollout Of Consumers
Google is replacing Gemini 2.5 Flash with Gemini 3 Flash as the default model in the Gemini app worldwide. The Pro model is still available for users to select from the model selector for coding and math problems.
According to the company, the new model is effective at recognizing multimodal content and providing you with a response based on it. For example, you can post a little video of yourself playing pickleball and get advice; you can attempt sketching and ask the model to guess what you’re drawing; or you may post an audio clip to receive feedback or create a test.
Additionally, according to the business, the model can produce richer visual responses with features like photos and tables and has a better understanding of the intent of users’ queries.
Additionally, you may utilize the new model to leverage prompts in the Gemini app to generate app prototypes.
Everyone in the United States can now search the Gemini 3 Pro, and more people can now access the Nano Banana Pro image model in search.
Availability Of Enterprises And Developers
The Gemini 3 Flash model, which is offered by Vertex AI and Gemini Enterprise, is already being used by organizations such as JetBrains, Figma, Cursor, Harvey, and Latitude, according to Google.
Through the API and Antigravity, Google’s new coding tool that was published last month, the corporation is providing developers with a preview model of the model.
According to the business, the Gemini 3 Pro outperforms GPT-5.2 with a score of 78% on the SWE-bench certified coding benchmark. It further stated that the model is perfect for data extraction, video analysis, and visual Q&A, and that its speed makes it appropriate for rapid and repeatable operations.
One million input tokens cost $0.50, whereas one million output tokens cost $3.00. This costs a little more than Gemini Flash 2.5’s $0.30 per million input tokens and $2.50 per million output tokens. However, Google asserts that the new model is three times faster and performs better than the Gemini 2.5 Pro model. Additionally, it typically utilizes 30% fewer tokens than 2.5 Pro for thinking tasks. In general, you may be able to reduce the quantity of tokens needed for specific tasks.
Flash is positioned as more of a workhorse model by us. From the standpoint of input and output prices, Flash is simply a far more affordable option if you consider, for instance, the input and output prices at the top of this table. Tulsee Doshi, senior director and head of product at Gemini Models, told during a briefing that “it actually allows for, for many companies, bulk tasks.”
In the midst of its intense release and performance fight with OpenAI, Google has processed over 1 trillion tokens daily on its API since the introduction of Gemini 3.
Sam Altman allegedly sent the OpenAI team an internal “Code Red” message earlier this month when ChatGPT’s traffic decreased while Google’s consumer market share increased. Following that, OpenAI launched a new picture generation model with GPT-5.2. Additionally, OpenAI boasted about its expanding enterprise utilization, claiming that since November 2024, the amount of ChatGPT messages has increased eightfold.
Google stated that the introduction of new models is making it difficult for all businesses to remain active, even though it did not specifically address the competition with OpenAI.
The way the business is going, it seems like all of these models are still amazing, pushing the boundaries and challenging one another. Additionally, I believe it’s fantastic that businesses are introducing these models,” Doshi stated.
Additionally, we are offering new metrics and methods for assessing these models. Thus, that also gives us encouragement.