Google just rolled out Gemini 3 Flash, a faster and cheaper version built on the Gemini 3 release from last month, and it’s becoming the default in the Gemini app and AI mode in search. This move seems aimed at shaking up the OpenAI competition by delivering a more affordable option without sacrificing performance.
But here’s where it gets interesting: Google claims Gemini 3 Flash not only outpaces its own 2.5 Flash variant but also matches or exceeds several frontier models in key tests. In Humanity’s Last Exam, for example, 3 Flash scored 33.7% without tool use, while 3 Pro hit 37.5%, 2.5 Flash was at 11%, and GPT-5.2 reached 34.5%. On the MMMU-Pro benchmark for multimodal reasoning, 3 Flash led all competitors with an 81.2% score. These numbers illustrate a notable leap in efficiency and capability, especially for deployments that rely on quick, multi-domain reasoning.
From a consumer perspective, Google is replacing Gemini 2.5 Flash as the default in the Gemini app, though users can still choose the Pro model for math and coding tasks. The company emphasizes that 3 Flash excels at multimodal content understanding and can generate answers that incorporate images, tables, and other visuals. Imagine uploading a short pickleball video and receiving tailored tips, sketching something and having the model identify what you drew, or uploading an audio clip for analysis or quiz generation.
The model is also positioned as better at inferring user intent and delivering more visual-rich responses. The Gemini app now enables features like app prototype creation from prompts, and the Pro variant remains available for broader search use.
Enterprise and developer access is already broadening. Google noted early adoption by JetBrains, Figma, Cursor, Harvey, and Latitude, with Gemini 3 Flash accessible via Vertex AI and Gemini Enterprise. Developers can preview the model through the API and in Antigravity, Google’s recent coding tool. In terms of performance, Gemini 3 Pro achieved 78% on the SWE-bench verified coding benchmark, with GPT-5.2 as its closest competitor. The company highlighted its suitability for video analysis, data extraction, and visual Q&A, touting faster speeds that support quick, repeatable workflows.
Pricing is $0.50 per 1 million input tokens and $3.00 per 1 million output tokens for Gemini 3 Flash Pro, which is slightly higher than the previous Gemini Flash 2.5 rates of $0.30 and $2.50, respectively. Google argues that 3 Flash outperforms 2.5 Pro while delivering about three times the speed, and for thinking tasks it uses roughly 30% fewer tokens on average than 2.5 Pro, potentially reducing total token counts for many tasks.
As Tulsee Doshi, Google’s senior director and head of Product for Gemini Models, explained, Flash is designed as the workhorse model: it’s a cost-efficient option optimized for bulk tasks where input and output costs add up. Since Gemini 3’s release, Google reports processing over 1 trillion tokens per day on its API, underscoring the intense competition in this space.
In response to Google’s push, OpenAI reportedly issued a “Code Red” internal memo as ChatGPT’s traffic growth slowed and Google’s consumer share rose. OpenAI has since released GPT-5.2 and a new image generation model, signaling a continued arms race across consumer, enterprise, and developer markets. While Google didn’t directly attack OpenAI, the company framed its approach as a challenge to all players to push the frontier and refine new evaluation benchmarks. Doshi emphasized that ongoing competition spurs improvements and new ways of measuring model capability.
Would you opt for Gemini 3 Flash as your primary workhorse for everyday tasks, even if it means paying a bit more per token for speed and multimodal features? How do you weigh price versus performance when choosing a foundation model for your business or personal use?
