Google rolls out Gemini 3.1 Flash-Lite preview for high-volume, cost-sensitive LLM workloads

Google announced Gemini 3.1 Flash-Lite on March 3, 2026 and started rolling it out in preview through the Gemini API in Google AI Studio and Vertex AI. Google positions the model as the fastest and most cost-efficient option in the Gemini 3 series for developers running high-volume workloads at scale.

What Google is shipping

According to Google, Gemini 3.1 Flash-Lite is priced at $0.25 per 1M input tokens and $1.50 per 1M output tokens. The company says the model improves on 2.5 Flash for both latency and throughput, citing Artificial Analysis benchmarks that show 2.5x faster time to first answer token and 45% higher output speed while maintaining similar or better quality.

Google also highlighted benchmark results including an Arena.ai Elo score of 1432, 86.9% on GPQA Diamond, and 76.8% on MMMU Pro. The company says the model includes configurable thinking levels in AI Studio and Vertex AI, letting developers choose how much reasoning depth they want for a given task.

Target workloads

Google is explicitly aiming Flash-Lite at translation, content moderation, labeling, and other recurring inference-heavy jobs where low latency and predictable cost matter. At the same time, the company is presenting the model as capable of more complex instruction following, with examples that include generating user interfaces, building dashboards, running simulations, and supporting multi-step business agents.

Early-access users cited by Google include Latitude, Cartwheel, Whering, and HubX. Those examples emphasize instruction following, multimodal labeling, and high-volume catalog or content pipelines rather than frontier reasoning research, which matches the model’s cost-first positioning.

Why it matters

The significance of this release is that Google is competing on operating economics as much as on raw model quality. For always-on systems such as search augmentation, moderation, or agent orchestration, price and speed often determine whether a feature can be deployed broadly. If the preview numbers translate into stable production behavior, Gemini 3.1 Flash-Lite gives teams another option for cost-sensitive LLM services.

Availability is still listed as preview, so the next test is operational rather than marketing: whether the model keeps the same latency, reliability, and quality profile under sustained real-world traffic and support commitments.

Source: Google