Gemini 3 Flash: Google's lightweight AI model stuns with 'raw speed' in AI Mode and more

What you need to know

• Google launched its Gemini 3 Flash model, bringing "raw speed" into the AI model world for developers and users alike.
• Gemini 3 Flash features "raw speed," outperforming the 2.5 Pro model at low costs: $0.50/1M input tokens and $3/1M output tokens.
• Gemini 3 Flash is available globally for all users, accessible in the Gemini app and becoming the "default model" for AI Mode in Search.

Gemini 3, Google's "most advanced" AI model family, is expanding this midweek, as the company debuts its smaller, faster variant globally.

This morning (Dec 17), Google launched Gemini 3 Flash, offering "frontier intelligence built for speed at a fraction of the cost." We've seen Flash models from Google before, and Gemini 3's version is no different, except for its power capabilities. The post states Gemini 3 Flash's main takeaway is that it features "raw speed" to outperform what was delivered through Gemini 2.5 Pro, with three times the quickness. Google states Gemini 3 Flash features the following pricing: $0.50/1M input tokens and $3/1M output tokens.

This speed is nothing without the artificial brains to back it up. Google claims the newest Flash model boasts PhD-level reasoning and "knowledge benchmarks like GPQA Diamond (90.4%) and Humanity’s Last Exam (33.7% without tools)."

More importantly, Gemini 3 Flash considers quality, as well as cost and speed, when thinking. Google says its AI model can "modulate" its thinking. While this may result in longer thinking times for "complex" tasks, Gemini 3 Flash still uses 30% fewer tokens than its hefty 2.5 Pro model.

The swift AI model also blazed through the MMMU Pro test, achieving high performance marks with a score of 81.2%.

Thinking in a Flash

Gemini 3 Flash: Frontier intelligence, built for speed - YouTube

Watch On

Speedy intelligence arrives for developers, too, as Gemini 3 Flash can reportedly outpace Google's Gemini 3 Pro and the 2.5 series with an SWE-bench Verified score of 78%. "Reasoning, tool use, and multimodal capabilities" were credited for the Flash model's prowess in complex video analysis.

Gemini 3 Flash is said to perform well for in-game assistants and A/B test experiments.

Google delivers another example, stating Gemini 3 Flash's enhanced multimodal capabilities can assist users in "seeing, hearing, and understanding" information quickly. Gemini can "understand" a video or image you've given it and create "useful" content from that understanding to further your agenda. Users can also create apps using your written prompt or voice without needing prior coding knowledge for something small and fun.

Gemini 3 Flash is available globally today (Dec 17) for everyone in the Gemini app. Google adds that this Flash model has become the "default model" in AI Mode for Search, too. Developers using the Gemini API can find 3 Flash in Google AI Studio, Gemini CLI, and Google Antigravity.

Enterprise users will see Gemini 3 Flash in Vertex AI and Gemini Enterprise.