Gemini 3.1 Flash-Lite: Built for intelligence at scale

Google DeepMind has slipped a new model into the Gemini 3 lineup, and this one is all about volume. Gemini 3.1 Flash-Lite is now rolling out in preview through AI Studio and Vertex AI, positioned as the model you reach for when you're running millions of requests and every fraction of a cent matters.

The pricing tells the story: $0.25 per million input tokens, $1.50 per million output tokens. That's Flash-Lite territory, but the performance numbers read more like a mid-tier model punching above its weight. According to Artificial Analysis benchmarks, it delivers a Time to First Answer Token 2.5 times faster than Gemini 2.5 Flash, with output speed up 45%. On the Arena.ai leaderboard it posted an Elo of 1432, and it scored 86.9% on GPQA Diamond and 76.8% on MMMU Pro — numbers that, notably, edge out 2.5 Flash, a model from a supposedly higher tier just one generation back.

What's more interesting than the raw scores is the control Google is handing developers. Flash-Lite ships with adjustable thinking levels baked into AI Studio and Vertex AI, letting teams dial reasoning effort up or down depending on the job. Cheap, low-effort inference for bulk translation or content moderation is one mode. Flip the switch and the same model can handle heavier lifting — generating dashboards, running simulations, following multi-step instructions — without needing to swap to a pricier tier.

Early adopters like Latitude, Cartwheel, and Whering are already testing it in production, and Google says feedback points to a model that handles complex inputs with the precision typically reserved for larger, more expensive systems, while sticking closely to instructions. That's the pitch, anyway: less a stripped-down budget option and more a scaling tool that happens to be cheap.