Google Launches Gemini 3 Flash - Now Default in Gemini App

Google Launches Gemini 3 Flash Model

Google has announced the release of Gemini 3 Flash, a new AI model designed for speed and cost-effectiveness. This launch follows the initial release of the broader Gemini 3 model last month, and is strategically positioned to compete with offerings from OpenAI.

The company is integrating Gemini 3 Flash as the standard model within both the Gemini application and the AI-powered features available in Google Search.

Significant Improvements Over Previous Models

The introduction of the Flash model comes half a year after the unveiling of Gemini 2.5 Flash. It represents a substantial advancement in capabilities and efficiency.

Testing demonstrates that Gemini 3 Flash significantly surpasses the performance of its predecessor. In certain evaluations, it achieves parity with leading models such as Gemini 3 Pro and GPT-5.2.

Performance Benchmarks

Gemini 3 Flash achieved a score of 33.7% on Humanity’s Last Exam, a comprehensive benchmark assessing knowledge across diverse fields, when evaluated without tool usage.

For comparison, Gemini 3 Pro attained 37.5%, Gemini 2.5 Flash scored 11%, and GPT-5.2 reached 34.5% on the same benchmark.

Leading in Multimodal Reasoning

The new model excelled in multimodal and reasoning tasks, as measured by the MMMU-Pro benchmark.

Gemini 3 Flash achieved a leading score of 81.2% on MMMU-Pro, exceeding the performance of all other competing models.

These results highlight the model’s enhanced ability to process and understand complex information from multiple sources.

Gemini 3 Flash Now Available to All Users

Google has initiated a global rollout, establishing Gemini 3 Flash as the standard model within the Gemini application. This update supersedes the previous Gemini 2.5 Flash model.

Despite this change, users retain the ability to select the Pro model via the model selection tool, particularly for tasks involving mathematical problems and coding inquiries.

Enhanced Multimodal Capabilities

According to Google, the newly implemented model demonstrates improved proficiency in recognizing and interpreting multimodal content. This allows for more relevant and insightful responses.

For example, users can now upload short videos, such as pickleball gameplay, and request advice. Alternatively, a simple sketch can be submitted for the model to attempt identification.

Furthermore, audio recordings can be uploaded for analysis or to automatically generate quizzes.

Improved Understanding and Visual Responses

The company highlights that the model exhibits a greater understanding of user intent when processing queries.

This enhanced comprehension translates into the generation of more visually engaging responses, incorporating elements like images and tables to better convey information.

App Prototyping Feature

Gemini 3 Flash also introduces the capability to create application prototypes directly within the Gemini app, all driven by text-based prompts.

Expanded Access to Gemini Pro and Nano

The Gemini 3 Pro model is now accessible to all users within the United States for use with search functionalities.

In addition, access to the Nano Banana Pro image model within search has been broadened to include a larger user base in the U.S.

Enterprise and Developer Access to Gemini 3 Flash

Google has announced that several organizations, including JetBrains, Figma, Cursor, Harvey, and Latitude, are currently utilizing the Gemini 3 Flash model. Access is provided through both Vertex AI and Gemini Enterprise platforms.

Developers can now experiment with the model via a preview offering accessible through the API. Furthermore, it’s integrated into Antigravity, Google’s recently launched coding environment.

Performance and Capabilities

According to Google, Gemini 3 Pro achieves a score of 78% on the SWE-bench verified coding benchmark. This performance is surpassed only by GPT-5.2. The model demonstrates particular strengths in areas such as video analysis, data extraction, and visual question answering.

Its speed makes it exceptionally well-suited for workflows requiring rapid and repetitive execution.

Pricing Details

The cost for utilizing the model is set at $0.50 per 1 million input tokens and $3.00 per 1 million output tokens. This represents a slight increase compared to the $0.30 and $2.50 rates, respectively, for Gemini Flash 2.5.

However, Google asserts that the new model delivers superior performance to Gemini 2.5 Pro while operating at three times the speed. Moreover, it typically consumes 30% fewer tokens for complex reasoning tasks than its predecessor, potentially leading to cost savings overall.

Positioning and Use Cases

“We view Flash as a robust workhorse model,” explained Tulsee Doshi, Senior Director & Head of Product for Gemini Models, during a briefing with TechCrunch. “Considering the input and output pricing, Flash offers a more economical solution, particularly for companies handling large volumes of tasks.”

The model’s affordability enables efficient processing of bulk operations.

Industry Context and Competition

Since the release of Gemini 3, Google’s API has processed over 1 trillion tokens daily, reflecting an intense competitive landscape with OpenAI.

Reports indicate that Sam Altman, CEO of OpenAI, issued an internal “Code Red” memo to his team following a decline in ChatGPT traffic coinciding with Google’s increased market share. In response, OpenAI has launched GPT-5.2 and a new image generation model, also highlighting growth in enterprise adoption and an eightfold increase in ChatGPT message volume since November 2024.

While refraining from direct commentary on OpenAI’s actions, Google emphasized that the introduction of new models fosters innovation and encourages continuous improvement across the industry.

“The ongoing developments within the industry are driving all companies to remain proactive and innovative,” Doshi stated. “The release of these models is also prompting the development of new benchmarks and evaluation methods, which is a positive trend.”