Overview
Gemini 3.1 Flash-Lite is the stable release of the model that first appeared in preview on March 3, 2026. As of May 2026 it is available as the production model identifier gemini-3.1-flash-lite, replacing the earlier gemini-3.1-flash-lite-preview. Google positions it as its fastest and most cost-efficient model in the Gemini 3 series, built for high-volume developer workloads at scale.
The model targets latency- and cost-sensitive tasks such as translation, classification, structured extraction, simple data processing, and lightweight agentic workflows — cases where per-token cost and response speed matter more than the heaviest multi-step reasoning. It complements Gemini 3 Pro and Gemini 3 Flash as the lowest-cost, high-throughput option in the family.
What's New
Stable model identifier
The release moves from the experimental preview label to the stable gemini-3.1-flash-lite identifier in the Gemini API. Teams that integrated the preview should update to the stable identifier and re-run quality, latency, and cost checks against their own token mix, since model behavior and pricing tiers are now finalized.
Confirmed pricing structure
The stable model has a published Gemini API pricing structure with a free tier and several paid inference modes (Standard, Batch, Flex, Priority), giving teams explicit cost options for high-volume deployment rather than the single preview rate.
Official positioning and benchmarks
Google's announcement frames Flash-Lite as the fastest, most cost-efficient Gemini 3 series model, supported by published benchmark results: an Elo score of 1432 on the Arena.ai Leaderboard, 86.9% on GPQA Diamond, 76.8% on MMMU Pro, plus 2.5X faster Time to First Answer Token and a 45% increase in output speed versus Gemini 2.5 Flash (the speed figures attributed by Google to Artificial Analysis benchmarking). It also includes thinking levels for adaptive reasoning control.
Performance Benchmarks
| Metric | Gemini 3.1 Flash-Lite | Source |
|---|---|---|
| Arena.ai Leaderboard (Elo) | 1432 | Google announcement |
| GPQA Diamond | 86.9% | Google announcement |
| MMMU Pro | 76.8% | Google announcement |
| Time to First Answer Token | 2.5X faster than Gemini 2.5 Flash | Google (Artificial Analysis) |
| Output speed | +45% vs Gemini 2.5 Flash | Google (Artificial Analysis) |
All figures are from Google's official announcement; the speed comparisons are attributed by Google to Artificial Analysis benchmarking. Results vary by workload and token mix.
Compatibility Notes
Gemini 3.1 Flash-Lite is available through the Gemini API in Google AI Studio and Vertex AI for enterprises, and is also offered as a model on Google Cloud's Gemini Enterprise Agent Platform. It supports multimodal input (text, image, video, and audio, with audio priced separately) and text output. Migrating from the preview only requires switching to the stable gemini-3.1-flash-lite identifier; validate behavior and cost against your own data before scaling.
Pricing & Plans
Gemini consumer access is free for general use, and the developer model is billed per token through the Gemini API with a free tier and four paid inference modes.
| Tier | Input (text/image/video) | Input (audio) | Output |
|---|---|---|---|
| Free | $0 | $0 | $0 |
| Standard | $0.25 / 1M | $0.50 / 1M | $1.50 / 1M |
| Batch | $0.125 / 1M | $0.25 / 1M | $0.75 / 1M |
| Flex | $0.125 / 1M | $0.25 / 1M | $0.75 / 1M |
| Priority | $0.45 / 1M | $0.90 / 1M | $2.70 / 1M |
The free tier is free of charge for both input and output across inference modes but excludes context caching, which is available on paid tiers at reduced rates. Confirm current rates on Google's official Gemini API pricing page before committing production budgets, since pricing can change.
Best For
- High-volume translation, classification, and structured-extraction pipelines that are cost-sensitive
- Latency-sensitive applications needing fast first-token response at low per-call cost
- Lightweight agentic workflows and simple data processing at scale
- Teams that prototyped on the March 2026 preview and need the stable model identifier
- Batch or asynchronous jobs that can use discounted Batch/Flex inference modes
FAQ
How is the stable release different from the March 2026 preview?
It moves from the experimental gemini-3.1-flash-lite-preview to the stable gemini-3.1-flash-lite model identifier with a finalized Gemini API pricing structure (free tier plus Standard, Batch, Flex, and Priority modes).
How much does Gemini 3.1 Flash-Lite cost?
There is a free tier. On the standard paid tier it is $0.25 per 1M text/image/video input tokens, $0.50 per 1M audio input tokens, and $1.50 per 1M output tokens, with cheaper Batch/Flex and higher Priority rates.
What is it best suited for?
High-frequency, lightweight tasks such as translation, classification, structured extraction, simple data processing, and lightweight agentic workflows where cost and latency are primary constraints.
Does it replace Gemini 3 Pro or Flash?
No. It complements them as the lowest-cost, highest-throughput option in the Gemini 3 series; heavier reasoning workloads may still need a higher-capability model.
Where can I access it?
Through the Gemini API in Google AI Studio and Vertex AI, and as a model on Google Cloud's Gemini Enterprise Agent Platform, using the stable gemini-3.1-flash-lite identifier.
