December 18th news, this morningGoogle Gemini 3 Flash was officially released, directly against the flagship models for OpenAI and Anthropic, officially called 2.5 Pro three times faster, with a price cut to one quarter of 3 Pro, and performance not reversed。
Starting today, users can experience Gemini 3 Flash (Fast/Thining) and Gemini 3 Pro in the Gemini product line。
The baseline test results showed that Gemini 3 Flash retained Pro-level reasoning capability but delayed and cost directly down to Flash。
Specifically, in a doctoral reasoning test like GPQA Diamond, it can get 904%, back and forth with the larger front models. The MMMU Pro test Gemini 3 Flash directly takes 81.2%, reaching the state-of-the-art level of industry, which is comparable to home-based 3Pro performance。
The data show that the Token consumption of Gemini 3 Flash is 30% less than 2.5 Pro, three times faster, and the price is down to the floor price of the input 0.5/million Token, and the output of $3/million Token。
In addition, Gemini 3 Flash has a bright spot on multi-modular capabilities, which can process visual, audio, etc. more quickly, and link "see, hear, understand" into a relatively smooth link, suitable for an interactive scene requiring instant feedback。
The test results show that Gemini 3 Flash not only fully transcends Gemini 2.5 Pro ' s multi-generational model, but is also able to resist the GPT-5.2 Chamber of OpenAI in several comparisons. The latter is the new model that OpenAI has urgently launched in response to Gemini 3 Pro。
In a difficult test like Humanity's Last Exam, both parties wereNo external tools such as searchI'm not sureLess than 1 percentage point difference.
Gemini 3 Flash Test, MMMMU-Pro, dedicated to assessing multimodular understanding and reasoning capabilities 81.2% ACHIEVEMENTS79.5% ABOVE GPT-5.2。
The results show that Gemini 3 Flash in "Extra High"GPT-5.2 in Extra High reasoning mode, the competitive pressure on OpenAI cannot be ignored。
ACCORDING TO 1AI, GOOGLE HASUse Gemini 3 Flash as the default modelTo Gemini App and AI Mode in search. Google stated that the model was open free to users around the world and brought about greater daily AI capabilities。
In addition, Google AI Mode added a direct access portal to Nano Banana Pro. Users in the U.S. region simply need to enable "Thining with 3 Pro" and "Create Images Pro" in the Model Selector to generate images in chat interfaces。