Gemini 3.1 Flash Lite: Google's Cheapest AI Model Yet

Google released Gemini 3.1 Flash Lite in preview on March 3, with general availability following on May 7, 2026, completing the 3.1 model family. Where Gemini 3.1 Pro was the family’s high-capability tier, Flash Lite is built for a different purpose: fast, cheap inference at the edge and in latency-sensitive applications.

Update (July 2026): the cheapest-tier crown has moved within Google’s lineup. Google shipped Gemini 3.5 Flash-Lite on July 21, 2026 at $0.30 per million input and $2.50 per million output, the fastest and cheapest model in the 3.5 line, although still pricier in absolute terms than 3.1 Flash Lite. Readers standardizing on the current 3.5 generation should start there; 3.1 Flash Lite remains the cheaper option outright while Google keeps it available. The reporting below reflects 3.1 Flash Lite as it stood at launch.

What Flash Lite Is For

Flash Lite occupies the smallest end of the Gemini model lineup. It’s designed for use cases where cost and speed matter more than maximum reasoning depth. Think mobile apps, embedded devices, high-volume classification tasks, and real-time features that need sub-second responses.

The model is smaller than Gemini 3.1 Flash. Google hasn’t published exact parameter counts for the production Flash models, but the difference shows up in inference speed and cost. Flash Lite runs significantly faster than Flash, which already runs faster than Pro.

Benchmark Performance

Flash Lite is built for a different job than Pro or Flash, so comparing it to those on capability benchmarks misses the point. Per Google’s model card, it posts 72.0% on LiveCodeBench (coding), 76.8% on MMMU-Pro (multimodal reasoning), and 88.9% on MMMLU (multilingual knowledge), with 86.9% on GPQA Diamond. These are adequate numbers for the use cases it’s targeting, not flagship-tier scores.

Where Flash Lite stands out is performance per dollar. For tasks like text classification, short-form summarization, information extraction, and simple question answering, a well-prompted Flash Lite will handle most inputs correctly at a fraction of what a Pro model costs. At production scale, that difference adds up fast.

Pricing and API Access

Gemini 3.1 Flash Lite is the cheapest model in the Gemini 3.1 family. At current rates, input tokens run $0.25 per million and output tokens $1.50 per million, well below standard Gemini 3.1 Flash at $1.50 per million input and $9.00 per million output, and a fraction of Pro. These prices make Flash Lite viable for applications that would be cost-prohibitive with larger models.

The model is available through the Gemini API and in Google AI Studio. For Android developers, on-device deployment is an option through Google’s AI Edge SDK, which lets you run Flash Lite locally without network latency or per-query API costs.

On-Device and Edge Deployment

The on-device story is where Flash Lite gets more interesting. Running inference locally means no round-trip latency, no per-query cost, and no data leaving the device. For consumer apps that need AI features like smart replies, local content moderation, or offline assistance, that shifts the cost calculation entirely.

Flash Lite fits in the emerging category of models small enough to run on high-end mobile hardware. A flagship Android phone or iPad can run Flash Lite inference at reasonable speeds without draining the battery in a session. That’s a threshold that larger Flash and Pro models don’t meet.

Where It Fits Against Competitors

The lightweight model category has several strong players. Meta’s Llama 3 8B, Microsoft’s Phi-3 Mini, and Apple’s on-device models all compete in this space. Flash Lite’s advantage is Gemini’s multimodal training: it understands images and text natively, which the smallest Llama variants don’t do as well.

For developers building on Google’s platform, Flash Lite is the natural starting point for any feature where you want AI capability without paying Pro prices. The sensible approach is to build with Flash Lite first, identify the cases where it falls short, and route only those to larger models.

When to Use It

Flash Lite is right for three situations: high-volume tasks where you need to run millions of queries per day and cost is a constraint; real-time features where response time under 200ms matters; and on-device deployments where network access isn’t reliable or data privacy requires local processing.

It’s wrong for complex reasoning, multi-step problem solving, long document analysis, and any task where getting the answer right on the first try is more important than getting it quickly or cheaply. The standard rule applies: match the model to the task, and don’t pay for capability you don’t need.

Gemini 3.1 Flash Lite rounded out the 3.1 model family. With 3.1 Pro at the top, 3.1 Flash in the middle, and Flash Lite for edge and volume work, the 3.1 generation spanned the full range from heavy reasoning to lightweight real-time apps. Google has since moved to the 3.5 generation, with Gemini 3.5 Flash going generally available at Google I/O in May 2026 as the new default model across the Gemini app and AI Mode in Search.