Gemini 3.1 Flash Lite

Supported Functionality

Item	Specification
Input	Text, image, video, audio, PDF
Output	Text
Context	1,048,576 tokens
Max Output	65,536 tokens
Vision	✓ Supported
Function Calling	✓ Supported

Description

Gemini 3.1 Flash-Lite is Google's lightweight, natively multimodal reasoning model in the Gemini 3 family. Its preview launched on March 3, 2026, followed by the stable gemini-3.1-flash-lite endpoint on May 7, 2026. Based on Gemini 3 Pro, it targets high-throughput, latency-sensitive, and cost-sensitive tasks, accepts text, image, video, audio, and PDF inputs, and has a January 2025 knowledge cutoff. Google has not disclosed its parameter count.

Its main advance is bringing adjustable reasoning and native multimodal processing to a lightweight price tier. Google reports high-thinking scores of 86.9% on GPQA Diamond, 76.8% on MMMU-Pro, and 72.0% on LiveCodeBench. Citing Artificial Analysis, Google also reports 2.5 times faster time to first answer token and 45% higher output speed than Gemini 2.5 Flash. The stable endpoint has an earliest shutdown date of May 7, 2027, with gemini-3.5-flash-lite listed as the recommended replacement.

Key Capabilities

• Efficient Reasoning: Supports minimal, low, medium, and high thinking levels, allowing developers to balance latency for bulk tasks against deeper reasoning for harder requests.
• Native Multimodal Understanding: Jointly analyzes text, images, video, audio, and PDFs, scoring 76.8% on MMMU-Pro for cross-media classification, labeling, and extraction.
• Million-Token Context: Accepts up to 1,048,576 tokens for large document sets, long recordings, video, and cross-file synthesis, although detailed recall at extreme lengths should be evaluated per workload.
• Lightweight Agents: Function calling, code execution, structured outputs, and URL Context support query routing, data processing, and multi-step business automation.
• Grounded Retrieval: Google Search, Google Maps, and File Search grounding enable responses informed by current web information, location data, and private corpora.
• Multilingual Processing: An 88.9% MMMLU score supports high-volume translation, moderation, customer-message classification, and cross-language question answering.
• Fast Code Generation: A 72.0% LiveCodeBench score supports UI, dashboard, and routine script generation, while complex software engineering is better assigned to a higher-tier model.

Technical Strengths

Feature	Benefit
Low-Latency Inference	Google's measured output speed of 363 tokens per second helps reduce interactive wait times and raise batch throughput.
Adjustable Thinking Levels	Developers can match reasoning depth to task difficulty for finer control over quality, latency, and cost.
Native Multimodal Architecture	A unified model reduces the need for separate image, audio, and video recognition pipelines in cross-media applications.
Very Large Input Capacity	The million-token window reduces fragmentation of large corpora and supports synthesis across sections and files.
Broad Tool Ecosystem	Function calling, grounded search, file retrieval, and code execution connect language understanding to real business operations.
Scalable Inference Options	Batch, Flex, and Priority inference let teams optimize for offline throughput, elastic capacity, or prioritized latency.