Google launches Gemini 3 Flash

Google has announced Gemini 3 Flash, its latest fast and affordable AI model, which will become the default model in the Gemini app and in the Search UI. The upgrade comes just six months after the release of Gemini 2.5 Flash, and brings significant improvements in speed, multimodal capabilities, and understanding of user intent. Google aims to directly compete with OpenAI, which recently released GPT‑5.2.

Gemini 3 Flash is based on the Gemini 3 architecture and achieves results comparable to the most advanced models in several key tests. On test Humanity's Last Exam It achieved 33.7 %, which is almost three times more than its predecessor (11 %), and comparable to the GPT‑5.2 (34.5 %) and Gemini 3 Pro (37.5 %).

On the MMMU-Pro multimodal test, which tests the understanding of images, text, sound and other data, the new model surpassed all competitors with an outstanding score of 81.2 %.

Default model for users worldwide

Gemini 3 Flash will now be the default model in Gemini, replacing version 2.5 Flash. Users can still switch to the Pro model when they need more advanced math or programming solutions.

The new model offers better interpretation of multimodal content and enables tasks such as:

analysis of the user's video,
interpretation of sketches or drawings,
processing of audio recordings,
generating quizzes or audio-based content,
preparation of application prototypes.

Gemini 3 Flash better understands user intent, creates more visually rich responses (tables, images), and enables improved tracking of elements across multiple creation steps.

For businesses and developers

Companies like JetBrains, Figma, Cursor, Harvey, and Latitude are already using the new model through Vertex AI and Gemini Enterprise. It is available to developers in an API preview and in Google's new Antigravity programming tool.

Google says that the Gemini 3 Pro scores 78 % on the SWE-bench verified encoding test, second only to GPT-5.2. The new Flash model is particularly well-suited for video analysis, data extraction, and fast, repeatable workflows due to its speed.

Prices:

$0.50 per 1 million input tokens,
$3.00 per 1 million output tokens.

The model is slightly more expensive than the Gemini 2.5 Flash, but Google claims it is three times faster and uses 30 % fewer tokens on "thinking" tasks, which can reduce overall costs.

Google and OpenAI are increasingly fierce rivals

Google processes over 1 trillion tokens per day through Gemini APIs as the race with OpenAI intensifies. According to recent reports, Sam Altman declared an internal “Code Red” earlier this month after ChatGPT’s traffic declined while Gemini’s user base grew.

Google does not officially comment on the "match results", but acknowledges that the rapid release of new models in the industry "encourages all players to innovate" and to develop new methods for assessing the quality of models.