Google today dropped its fast and cost-effective Gemini 3 Flash model, based on last month’s Gemini 3, clearly aiming to rival OpenAI. The company is making this the default model in the Gemini app and AI mode in search, though users can still pick the Pro model for math or coding tasks.
Arriving six months after Gemini 2.5 Flash, this new model shows great improvements. Benchmarks suggest it outperforms its predecessor and, in some areas, even matches high-end models like Gemini 3 Pro and GPT-5.2. For example, it scored 33.7% without tool use on the Humanity’s Last Exam benchmark compared with Gemini 3 Pro at 37.5% and GPT-5.2 at 34.5%. On the MMMU-Pro reasoning benchmark, it beat competitors with 81.2%. Gemini 3 Flash shines at multimodal understanding. Users can upload videos, sketches, or audio for analysis, tips, or quizzes. It also delivers answers with images and tables, and can even help build app prototypes directly in the Gemini app.
Also Read| IRCTC's new user numbers see massive change as crackdown on fake IDs takes effect
Enterprise adoption and developer access
Enterprise adoption is already underway. Companies like JetBrains, Figma, Cursor, Harvey, and Latitude are using the model via Vertex AI and Gemini Enterprise. Developers can try it in preview via the API or Google’s Antigravity coding tool.
Also Read| Your AI chats are going on sale! Did you share too much?
Pricing and industry impact
Pricing is USD 0.50 per million input tokens and USD 3 per million output tokens, slightly higher than 2.5 Flash, but the model is three times faster and uses 30% fewer tokens for thinking tasks. Tulsee Doshi, senior director of Gemini Models, called it a “workhorse model” built for bulk and repeatable workflows. Since Gemini 3’s launch, Google has processed over 1 trillion tokens per day on its API. While not calling out OpenAI directly, the company says new models continue to push the frontier and encourage innovation.
