Google has launched a new AI model called Gemini 3.1 Flash-Lite. The company announced this on March 3, 2026. Google describes it as the fastest and most cost-efficient model in the Gemini 3 series.
Developers can now access this model in preview. They use it through the Gemini API in Google AI Studio. Enterprises access it via Vertex AI. However, end users cannot try it yet.
Google highlights strong improvements over previous versions. The model delivers a 2.5 times faster time to first token. It also offers 45% higher output speed compared to Gemini 2.5 Flash. Benchmarks from Artificial Analysis support these claims. On the Arena.ai leaderboard, it scores an Elo of 1432. Google states that it beats models like GPT-5 mini, Claude 4.5 Haiku, and Grok 4.1 Fast in output speed.
The model supports two modes in AI Studio and Vertex AI. Developers choose standard mode or thinking mode. In thinking mode, users control how much time the model spends thinking about a task.
Google points to many practical uses. Developers apply it for high-volume translation. It handles content moderation well. The model also generates user interfaces and dashboards. It creates simulations. It follows complex instructions effectively.
Pricing makes this model very attractive. Google charges $0.25 for one million input tokens. Output costs $1.50 per million tokens. In comparison, Gemini 2.5 Flash costs more at $0.30 for input and $2.50 for output per million tokens.
This launch shows Google’s focus on efficient AI. Developers now get powerful tools for large-scale work without high costs. The preview phase lets teams test and build with it right away.