Top AI Product

Google quietly dropped [Gemini 3.1 Flash-Lite](https://deepmind.google/models/model-cards/gemini-3-1-flash-lite/) on March 3rd, and honestly, the pricing alone made me do a double-take. We’re talking $0.25 per million input tokens and $1.50 per million output tokens. That’s one-eighth the cost of Gemini 3.1 Pro. For bulk workloads like translation, content moderation, or e-commerce processing, those numbers start to matter a lot.

But cheap doesn’t mean slow. Google claims 2.5x faster time-to-first-token compared to Gemini 2.5 Flash, plus a 45% bump in output speed. I’ve been poking around with it in [Google AI Studio](https://ai.google.dev/gemini-api/docs/changelog), and the response times genuinely feel snappy — even with reasoning turned on.

Speaking of reasoning, the standout feature here is adjustable thinking levels. You get four options: minimal, low, medium, and high. So if you’re running a simple classification task, you can dial it down to minimal and save on latency. Need something more thoughtful? Crank it up. [Simon Willison tested this out](https://simonwillison.net/2026/Mar/3/gemini-31-flash-lite/) on his blog by having the model generate pelican drawings at each thinking level — a fun way to visualize the difference.

The buzz has been real. It picked up [158 upvotes on Product Hunt](https://www.producthunt.com/products/gemini-6) on March 4th, [VentureBeat ran a piece](https://venturebeat.com/technology/google-releases-gemini-3-1-flash-lite-at-1-8th-the-cost-of-pro) highlighting the 1/8th cost angle, and [MarkTechPost](https://www.marktechpost.com/2026/03/03/google-drops-gemini-3-1-flash-lite-a-cost-efficient-powerhouse-with-adjustable-thinking-levels-designed-for-high-scale-production-ai/) did a deep breakdown of the architecture. Over on [Hacker News](https://news.ycombinator.com/item?id=47234962), folks were already comparing it against other budget models, with one commenter noting that “Gemini is slowly making $15/month voice apps obsolete.”

The benchmarks back up the hype too. Google reports an Elo score of 1432 on the Arena.ai leaderboard and 86.9% on GPQA Diamond, which puts it ahead of its predecessor across key reasoning and multimodal tasks. For a model at this price point, that’s impressive.

If you’re running anything at scale where cost-per-token keeps you up at night, Flash-Lite is worth a serious look. It’s currently available in preview through the [Gemini API](https://ai.google.dev/gemini-api/docs/changelog) and [Vertex AI](https://docs.cloud.google.com/vertex-ai/generative-ai/docs/models/gemini/3-1-flash-lite). Not the flashiest launch Google has ever done, but maybe the most practical one in a while.