Google has launched its fast and cost-effective Gemini 3 Flash model today, building on the Gemini 3 series unveiled last month, in a strategic move to capture market share from OpenAI. The company is now setting this new model as the default for AI Mode across both the Gemini app and Google Search.
The release of the new Flash model comes just six months after Google introduced Gemini 2.5 Flash, marking a significant performance leap. In benchmark evaluations, Gemini 3 Flash outperforms its predecessor across multiple metrics and even rivals top-tier models like Gemini 3 Pro and GPT-5.2 in certain areas.
For instance, it achieved a score of 33.7% on the HumanEvalPlus benchmark, which assesses expertise across diverse domains. This compares to 37.5% for Gemini 3 Pro, 11% for Gemini 2.5 Flash, and 34.5% for the recently released GPT-5.2.
In the multimodal and reasoning benchmark MMMU-Pro, the new model leads all competitors with an impressive score of 81.2%.
Consumer Rollout
Google is making Gemini 3 Flash the default model globally within the Gemini app, replacing Gemini 2.5 Flash. Users can still opt for the Pro model via the model selector when handling complex math or coding tasks.
The company highlights that the updated model excels at interpreting multimodal inputs and generating context-aware responses. For example, users can upload a short pickleball video clip to receive playing tips, sketch a rough drawing for the model to identify, or submit audio recordings for analysis and quiz generation.
Additionally, the model demonstrates improved understanding of user intent and produces more visually engaging outputs, including images and formatted tables.
With the new model, users can also prototype applications directly in the Gemini app using simple prompts. Meanwhile, Gemini 3 Pro is now widely available in the U.S. for search, and more American users have access to the Nano Banana Pro image model within search features.
Enterprise and Developer Availability
Google noted that companies such as JetBrains, Figma, Cursor, Harvey, and Latitude are already integrating Gemini 3 Flash into their workflows. The model is accessible through Vertex AI and Gemini Enterprise platforms.
For developers, Google offers preview access via API and Antigravity, its newly launched coding assistant introduced last month.
Gemini 3 Pro scored 78% on the SWE-bench validation benchmark for code generation, placing it just behind GPT-5.2. Google emphasizes that the model is well-suited for video analysis, data extraction, and visual question answering, and thanks to its speed, supports rapid and scalable workflows.
Pricing for Gemini 3 Flash is set at $0.50 per million input tokens and $3.00 per million output tokens—slightly higher than Gemini 2.5 Flash, which was priced at $0.30 and $2.50 respectively. However, Google asserts that the new model surpasses the performance of Gemini 2.5 Pro while being three times faster. Moreover, it uses 30% fewer tokens on average for reasoning tasks compared to 2.5 Pro, potentially reducing overall token consumption and costs in practice.
"We really position Flash as more of a workhorse model. If you look at the input and output pricing at the top of this table, Flash is actually a more economical option for both inputs and outputs. This enables many businesses to run large-scale batch operations efficiently," said Tulsee Doshi, Senior Director of Product for Gemini models, in a briefing with TechCrunch.
Since launching Gemini 3, Google has been processing over one trillion tokens daily via its APIs, amid an intense race with OpenAI marked by frequent releases and performance benchmarks.
Earlier this month, Sam Altman reportedly issued a "red code" internal memo to OpenAI staff due to declining ChatGPT traffic and rising consumer adoption of Google's AI offerings. In response, OpenAI has since released GPT-5.2 and a new image generation model. It has also highlighted growth in enterprise usage, claiming an eightfold increase in ChatGPT message volume since November 2024.
While Google did not explicitly address the rivalry with OpenAI, it emphasized that continuous model innovation keeps the entire industry competitive.
"What's happening across the industry is that these models keep getting better, pushing each other forward and advancing the frontier. It's equally exciting how companies respond when new models are released," Doshi added.
"We're also introducing new benchmarks and novel ways to evaluate these models—which in turn motivates us to keep innovating."
