Google today released the fast and cheap Gemini 3 Flash model, based on last month’s Gemini 3, in an attempt to steal OpenAI’s thunder. The company also makes this the default model in the Gemini app and the AI feature in search.
The new Flash model comes six months after Google announced the Gemini 2.5 Flash model, offering significant improvements. In the benchmark, the Gemini 3 Flash model outperforms its predecessor by a significant margin and matches the performance of other flagship models such as the Gemini 3 Pro and GPT 5.2 on some measures.
For example, he scored 33.7% without using a tool The benchmark for Humanity’s Last Examination, which is designed to test expertise in different areas. 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 MMMU-Pro multimodality and reasoning benchmark, the new model outperformed all competitors with a score of 81.2%.
Disposition to consumers
Google makes Gemini 3 Flash the default model in the Gemini app worldwide, replacing Gemini 2.5 Flash. Users can still select the Pro model from the model selector for math and coding questions.
The company says the new model is good at recognizing multimodal content and giving you a response based on that. For example, you can upload the short pickleball video and ask for tips. You can try to draw a sketch and ask the model to guess what you are drawing. or you can upload a recording to get analysis or create a quiz.
The company also said the model better understands the intent of users’ queries and can generate more visual responses with elements such as images and tables.
Techcrunch event
San Francisco
|
13-15 October 2026
You can also use the new model to create app prototypes in the Gemini app using prompts.
Gemini 3 Pro is now available to everyone in the US to search, and more people in the US can also access the Nano Banana Pro image model in search.
Business and developer availability
Google noted that companies such as JetBrains, Figma, Cursor, Harvey and Latitude are already using the Gemini 3 Flash model, which is available through Vertex AI and Gemini Enterprise.
For developers, the company is making the model available in a preview model via the API and in Antigravity, Google’s new coding tool released last month.
The company said the Gemini 3 Pro scores 78% on the SWE-bench verified encoding benchmark, bested only by the GPT-5.2. He added that the model is ideal for video analysis, data mining and visual Q&A, and because of its speed, it is suitable for fast and repetitive workflows.
Model pricing is $0.50 per 1 million input tokens and $3.00 per 1 million output tokens. This is slightly more expensive than Gemini Flash 2.5’s $0.30 per 1 million input tokens and $2.50 per 1 million output tokens. But Google claims the new model outperforms the Gemini 2.5 Pro while being three times faster. And, for think tasks, it uses 30% fewer tokens on average than 2.5 Pro. This means that overall, you may save the number of tokens for some tasks.
“We’re really positioning flash as more of your work model. So if you look at, for example, even the input and output prices at the top of this table, Flash is just a much cheaper offering from an input and output price point of view. And so it actually enables, for a lot of companies, bulk jobs,” he told a.
Since the launch of Gemini 3, Google has been processing over 1 trillion tokens per day in its API amid a fierce traffic and performance war with OpenAI.
Earlier this month, Sam Altman reportedly sent an internal “Code Red” note to the OpenAI team afterwards ChatGPT traffic is down as Google’s market share among consumers increased. After that, OpenAI released GPT-5.2 and a new image generation model. OpenAI also boasted about its growing business usage and said that the volume of ChatGPT messages has increased 8x since November 2024.
While Google didn’t directly address the competition with OpenAI, it said the release of new models challenges all companies to be active.
“What’s happening across the industry is like all these models are continuing to be awesome, challenging each other, pushing boundaries. And I think what’s also awesome is as companies are releasing these models,” Doshi said.
“We’re also introducing new benchmarks and new ways to evaluate these models. And that’s encouraging as well.”
