Google today aims to steal OpenAI’s thunder with the release of its faster and cheaper Gemini 3 Flash model, which is based on Gemini 3 released last month. The company has also made this the default model for the Gemini app and AI mode in search.
The new Flash model comes six months after Google announced the Gemini 2.5 Flash model, and it comes with significant improvements. In benchmarks, the Gemini 3 Flash model significantly outperforms its predecessor and in some respects matches the performance of other Frontier models such as the Gemini 3 Pro and GPT 5.2.
For example, on Humanity’s Last Exam benchmark, which is designed to test expertise across a variety of domains, I scored 33.7% without using the tool. In comparison, Gemini 3 Pro had a score of 37.5%, Gemini 2.5 Flash had a score of 11%, and the newly released GPT-5.2 had a score of 34.5%.
In the multimodality and inference benchmark MMMU-Pro, the new model outperforms all competitors with a score of 81.2%.
Deployment to consumers
Google has adopted Gemini 3 Flash as the default model for Gemini apps worldwide, replacing Gemini 2.5 Flash. Users can still select Pro models from the model picker for math and coding questions.
The company says its new model is better at identifying multimodal content and providing answers based on it. For example, you can upload a short pickleball video and ask for tips. You can try drawing a sketch and ask the model to guess what you are drawing. Alternatively, you can also upload audio recordings to analyze and generate quizzes.
The company also said the model can better understand the intent of a user’s query and generate more visual answers using elements such as images and tables.
tech crunch event
san francisco
|
October 13-15, 2026
You can also use the new model to prototype your app in the Gemini app using prompts.
The Gemini 3 Pro is now available to everyone in the US in Search, and more people in the US can now also access the Nano Banana Pro image model in Search.
Enterprise and developer availability
Google said 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 this model available in preview via its API and in Google’s new coding tool, Antigravity, released last month.
According to the company, Gemini 3 Pro scores 78% on the SWE Bench-validated coding benchmark, with only GPT-5.2 outperforming it. He added that the model is ideal for video analysis, data extraction, and visual Q&A, and its speed makes it suitable for rapid, repeatable workflows.
The model is priced at $0.50 per million input tokens and $3.00 per million output tokens. This is slightly more expensive than Gemini Flash 2.5’s $0.30 per million input tokens and $2.50 per million output tokens. However, Google claims that the new model is three times faster than the Gemini 2.5 Pro model. Additionally, thinking tasks use an average of 30% less tokens than 2.5 Pro. This means that, overall, you may save on the number of tokens for a given task.
“We’re positioning flash as a flagship model. For example, if you look at the input and output prices at the top of this table, flash is just a much cheaper product from an input and output price perspective. So it actually enables bulk tasks for a lot of companies,” Tulsi Doshi, senior director and head of product at Gemini Models, told TechCrunch in a briefing.
Since releasing Gemini 3, Google has processed more than 1 trillion tokens per day with its API, amidst a fierce release and performance war with OpenAI.
Earlier this month, Sam Altman reportedly sent an internal “code red” memo to the OpenAI team after ChatGPT traffic declined as Google’s market share among consumers rose. OpenAI has since released GPT-5.2 and new image generation models. OpenAI also touted its growing adoption in enterprises, saying that ChatGPT’s message volume has increased eight times since November 2024.
Google did not directly address competition with OpenAI, but said the release of the new model requires all companies to be proactive.
“What’s happening across the industry is that all these models continue to be great and challenge each other and push the frontiers. And I think it’s also great that companies are releasing these models,” Doshi said.
“We are also introducing new benchmarks and new methods to evaluate these models, which is encouraging for us as well.”
