# Gemini 3.1 Flash Lite

Gemini 3.1 Flash Lite is the production release of the cost-efficient workhorse in the Gemini 3.1 family, with quality close to 2.5 Flash across key capability areas and a context window of 1M tokens for long-document workloads.

- **Model ID:** `google/gemini-3.1-flash-lite`
- **Type:** chat
- **Providers:** google, vertex
- **Context window:** 1,000,000
- **Maximum output tokens:** 65,000
- **Pricing:** $0.25/1M input tokens, $1.50/1M output tokens
- **Canonical page:** https://vercel.com/ai-gateway/models/gemini-3.1-flash-lite

## Supported parameters

Detailed capability metadata has not been reported for this model.

## Example

```ts
import { streamText } from 'ai'

const result = streamText({
  model: 'google/gemini-3.1-flash-lite',
  prompt: 'Why is the sky blue?'
})
```

## About

Gemini 3.1 Flash Lite is the general-availability version of the efficiency tier in the Gemini 3.1 generation, released May 7, 2026. It outperforms Gemini 2.5 Flash Lite on overall quality and lands close to 2.5 Flash performance across key capability areas, including reasoning, multimodal understanding, agentic tool use, and long-context performance.

The model is positioned for high-volume use cases where unit economics, not peak capability, set the constraint. Gemini 3.1 Flash Lite accepts text, images, audio, and documents as input within 1M tokens and produces text output, with implicit caching and web search available as runtime options to control cost and ground responses in current information.

Four configurable thinking levels (`minimal`, `low`, `medium`, `high`) allow a single deployment to serve mixed workloads without routing across models. A bulk extraction job can run at `minimal` to minimize latency and tokens; an edge case in the same pipeline can step up to `medium` or `high` when more deliberation pays off. Thinking tokens contribute to output token counts, so the level becomes a direct lever on cost.

Compared to the preceding preview release, Gemini 3.1 Flash Lite is the stable surface for production deployments. Teams running 2.5 Flash Lite at scale get the 3.1 generation quality gains on the workloads that drive the most tokens, including translation, data extraction, and code completion, without moving up to the standard Flash tier.

## What to consider

Gemini 3.1 Flash Lite exposes four thinking levels through `providerOptions.google.thinkingConfig`: `minimal`, `low`, `medium`, and `high`. Lower levels favor latency and per-token cost; higher levels add reasoning compute that counts toward output tokens. Benchmark total cost under realistic thinking settings before committing to a deployment.

## When to use

### Best For
- **High-volume agentic pipelines:** Aggregate token cost is a binding constraint and per-step reasoning depth can be tuned per request
- **Data extraction at scale:** Structured extraction over millions of documents, invoices, transcripts, or scraped HTML where throughput economics dominate
- **Bulk translation workloads:** Per-token cost determines whether the use case is viable, with thinking levels available for nuanced passages
- **Code completion and review:** Inline suggestions, lint-style review, and refactor proposals at IDE or CI/CD scale
- **Multimodal ingestion at the lite tier:** Vision and file inputs are required, but pro-tier latency and cost are not

### Consider Alternatives When
- **Maximum reasoning depth is required:** Complex multi-step problems benefit from `google/gemini-3.1-pro-preview` or `google/gemini-3-pro-preview`
- **Native image output is needed:** Gemini 3.1 Flash Lite returns text only; `google/gemini-3.1-flash-image-preview` or `google/gemini-3-pro-image` generate images
- **Pro-grade quality at flash latency:** `google/gemini-3-flash` sits between Flash Lite and Pro on capability and cost
- **Pure embedding workloads:** Semantic retrieval and clustering fit a dedicated embedding model like `google/gemini-embedding-001` better

## Best for

- **High-volume agentic pipelines:** Aggregate token cost is a binding constraint and per-step reasoning depth can be tuned per request
- **Data extraction at scale:** Structured extraction over millions of documents, invoices, transcripts, or scraped HTML where throughput economics dominate
- **Bulk translation workloads:** Per-token cost determines whether the use case is viable, with thinking levels available for nuanced passages
- **Code completion and review:** Inline suggestions, lint-style review, and refactor proposals at IDE or CI/CD scale
- **Multimodal ingestion at the lite tier:** Vision and file inputs are required, but pro-tier latency and cost are not

## Consider alternatives

- **Maximum reasoning depth:** Complex multi-step problems benefit from `google/gemini-3.1-pro-preview` or `google/gemini-3-pro-preview`
- **Native image output:** Gemini 3.1 Flash Lite returns text only; `google/gemini-3.1-flash-image-preview` or `google/gemini-3-pro-image` generate images
- **Pro-grade quality at flash latency:** `google/gemini-3-flash` sits between Flash Lite and Pro on capability and cost
- **Pure embedding workloads:** Semantic retrieval and clustering fit a dedicated embedding model like `google/gemini-embedding-001` better

## Frequently asked questions

### How is Gemini 3.1 Flash Lite different from `google/gemini-3.1-flash-lite-preview`?

Gemini 3.1 Flash Lite is the general-availability release of the same efficiency tier in the Gemini 3.1 family. The preview entry remains in the catalog for teams pinned to the earlier identifier; the GA model is the recommended target for new production work.

### How does Gemini 3.1 Flash Lite compare to Gemini 2.5 Flash Lite?

Gemini 3.1 Flash Lite outperforms 2.5 Flash Lite on overall quality and lands close to 2.5 Flash across reasoning, multimodal understanding, agentic tool use, and long-context performance. For teams already running 2.5 Flash Lite at scale, it's a quality upgrade within the same lite tier.

### What thinking levels does Gemini 3.1 Flash Lite support and how do they affect cost?

Four levels: `minimal`, `low`, `medium`, and `high`. Higher levels add reasoning compute that contributes to output token counts, so the choice trades off latency and per-request cost against quality on harder inputs.

### Can I mix thinking levels across requests in the same application?

Yes. Set `thinkingLevel` per request in `providerOptions.google.thinkingConfig`. Routine requests can run at `minimal` while flagged hard cases run at `medium` or `high` without any architectural changes.

### Does Gemini 3.1 Flash Lite support multimodal inputs?

Yes. Gemini 3.1 Flash Lite accepts text, images, audio, and documents as input within the 1M tokens context window and returns text output. Web search and implicit caching are available as runtime options.

### How do I call Gemini 3.1 Flash Lite on AI Gateway?

Use the identifier `google/gemini-3.1-flash-lite` with the AI SDK, the OpenAI-compatible Chat Completions endpoint, the Responses API, or any other supported interface. AI Gateway handles provider routing, retries, and failover automatically.

### How does Zero Data Retention work with Gemini 3.1 Flash Lite through AI Gateway?

Yes, Zero Data Retention is available for this model. Zero Data Retention is offered on a per-provider basis. See https://vercel.com/docs/ai-gateway/capabilities/zdr for details.

## Links

- [Model page](https://vercel.com/ai-gateway/models/gemini-3.1-flash-lite)
- [AI Gateway documentation](https://vercel.com/docs/ai-gateway)
- [Provider model documentation](https://ai.google.dev/gemini-api/docs/models)