Loading page
LightweightGoogle
Gemini 3.5 Flash-Lite
Gemini 3.5 Flash-Lite is a lightweight-tier multimodal model from Google's Gemini family, with a roughly 1M-token context window and support for text, image, video, and audio inputs.
Context 1.0M
Tier Lightweight
Tools Supported
Modalities text, image, video, audio
Input from
$0.150 / 1M tokens
across 3 providers
API Pricing
Cheapest on Google Cloud — 38% below avg| Provider | Input / 1M | Output / 1M | Cached / 1M |
|---|---|---|---|
| $0.150 | $1.25 | $0.015 | |
| $0.150 | $1.25 | $0.015 | |
| $0.300 | $2.50 | $0.030 | |
| $0.300 | $2.50 | $0.030 | |
| $0.300 | $2.50 | $0.030 |
Prices updated daily. Last check: Sep 11, 2026
Gemini 3.5 Flash-Lite pricing by provider
Input, output, and batch rates, plus alternatives, for one provider at a time.
Performance & Benchmarks
Source: Artificial Analysis →Intelligence
22.7 / 100
Coding
49.3 / 100
Output Speed
350 t/s
Latency (TTFT)
7.9s
Reasoning & Knowledge
- GPQA Diamond83.8%
- Humanity's Last Exam18.8%
Coding
- SciCode41.3%
Agentic & Tool Use
- Terminal-Bench v2.153.6%
- τ-bench Banking17.5%
Instruction & Long Context
- Long-Context Reasoning76.0%
Benchmarks measured Sep 2026. Scores are independent evaluations, not vendor-reported.
Model Details
General
- Creator
- Family
- Gemini
- Tier
- Lightweight
- Context Window
- 1.0M
- Modalities
- Text, Image, Video, Audio
Capabilities
- Tool Calling
- Yes
- Open Source
- No
- Aliases
- gemini-3.5-flash-lite, Gemini 3.5 Flash-Lite, models/gemini-3.5-flash-lite
Strengths & Limitations
Strengths
- Approximately 1,048,576-token (1M) context window, allowing very long documents or transcripts in one request
- Accepts text, image, video, and audio inputs, covering media types many lightweight models do not
- Supports tool calling, enabling function invocation and agentic workflows at the lightweight tier
- Measured output throughput of about 350 tokens per second (Artificial Analysis), suiting streaming and high-volume generation
- Lightweight tier positioning makes it a candidate for the cost-sensitive, high-request-volume layer of a multi-model stack
- Long context plus native audio and video input suits media summarization and transcript analysis without separate preprocessing pipelines
Limitations
- Lightweight tier — not intended to match Gemini Flash or Pro tier siblings on complex reasoning and coding tasks
- Measured time to first token of roughly 7,494 ms in Artificial Analysis testing is high for an interactive, latency-critical use case
- Proprietary model served through APIs; weights are not distributed for self-hosting
- We do not track output modalities beyond text, so image or audio generation should not be assumed
- Effective use of the full 1M-token context depends on provider limits and increases per-request latency and cost
Key Features
•1,048,576-token context window
•Multimodal input: text, image, video, and audio
•Tool calling / function calling support
•Lightweight tier of the Gemini family
•Measured output speed of ~350 tokens/second (Artificial Analysis)
•Available under the model IDs gemini-3.5-flash-lite and models/gemini-3.5-flash-lite
•Long-context document and transcript processing
About Gemini 3.5 Flash-Lite
Gemini 3.5 Flash-Lite is a model in Google's Gemini family, positioned at the lightweight tier. Within the Gemini lineup, Flash-Lite variants sit below the Flash and Pro tiers and are oriented toward high-throughput, latency-sensitive workloads rather than the heaviest reasoning tasks. It is exposed through API identifiers such as `gemini-3.5-flash-lite` and `models/gemini-3.5-flash-lite`.
The model accepts a context window of 1,048,576 tokens (roughly one million), which allows long documents, transcripts, or large collections of retrieved passages to be processed in a single request. It is multimodal on the input side, handling text, image, video, and audio, and supports tool calling for function-invoking and agentic patterns. Independent measurements from Artificial Analysis put output throughput at roughly 350 tokens per second, with a measured time to first token of about 7.5 seconds in that test run — a combination of fast generation once streaming begins and a comparatively slow initial response in that measurement.
In practice, lightweight Gemini models are used where request volume is high and per-request cost and generation speed matter more than maximum reasoning depth: classification, extraction, summarization, transcript and media triage, and routing layers in multi-model systems. Compared with Flash or Pro tier siblings in the Gemini family, Flash-Lite is the lower-cost, higher-throughput option; workloads involving difficult multi-step reasoning or code generation are typically better matched to a higher tier.
Common Use Cases
Gemini 3.5 Flash-Lite fits workloads where throughput and cost per request dominate: large-scale classification and tagging, structured data extraction from documents, summarization of long reports or meeting transcripts, and first-pass triage in pipelines that escalate hard cases to a Flash or Pro tier model. Its multimodal input support makes it usable for describing or categorizing images, screening video and audio content, and captioning media at volume without a separate transcription step. The ~1M-token context window supports long-document question answering and retrieval-augmented setups with large retrieved contexts, while tool calling lets it act as the execution layer in agents whose planning is handled elsewhere. Given the measured time to first token, batch and asynchronous jobs are a better match than tight interactive loops.
Frequently Asked Questions
How much does Gemini 3.5 Flash-Lite cost?
Pricing varies by provider and by pricing type — input tokens, output tokens, cached input, and any media-specific rates are often billed differently, and providers update rates frequently. Check the pricing table on this page for current per-provider figures.
What is Gemini 3.5 Flash-Lite best used for?
It is aimed at high-volume, cost-sensitive tasks: classification, extraction, summarization, media triage, and routing within larger systems. Its ~1M-token context also makes it suitable for long-document and transcript processing, and tool calling allows it to serve as an execution step in agentic pipelines.
How does it differ from other Gemini tiers?
Flash-Lite is the lightweight tier of the Gemini family, positioned below Flash and Pro tier models. It trades reasoning depth for throughput and lower cost, so tasks requiring extended multi-step reasoning or complex code generation are usually better routed to a higher Gemini tier.
Can it process images, video, and audio?
Yes — our metadata confirms text, image, video, and audio inputs. That makes it usable for image classification, video screening, and audio or transcript summarization without a separate preprocessing model.
Is it fast enough for real-time chat?
Artificial Analysis measured output throughput of roughly 350 tokens per second, which is fast once generation starts, but a time to first token of about 7.5 seconds in the same test. For latency-critical interactive use, benchmark it on your own provider and prompt shape before committing.
Are the model weights available to download?
No. Gemini models are served through Google's API and partner platforms rather than distributed as downloadable weights, so deployment options are limited to hosted endpoints.