Google/gemini-3.1-flash-lite

Supported Functionality 表格 Item Specification Input Text, Image, Audio, Video, PDF Output Text Context 1,048,576 tokens Max Output 65,536 tokens Vision ✓ Supported (Image, Video, Scanned PDF & Document Receipt Parsing) Function Calling ✓ Supported (Parallel Batch Calling, Sandbox Code Execution, Web Grounding, Constrained Structured JSON, Prompt Caching) Description Paragraph 1: Released by Google DeepMind in March 2026, Gemini 3.1 Flash Lite is the most cost-effective and highest-throughput industrial-grade lightweight model in the Gemini 3.1 series, built on optimized sparse MoE architecture with knowledge cutoff in January 2025. It outperforms Gemini 2.5 Flash on mainstream benchmarks including GPQA and MMMU, designed for cost reduction in high-frequency mass API workloads targeting SMEs, consumer internet applications and enterprise bulk data processing scenarios. Paragraph 2: Optimized for lightweight computing based on Gemini 3.1 Flash, it retains native four-modal input, million-token context and adaptive multi-level reasoning. It delivers 2.5× faster time-to-first-token and a throughput of 363 tokens per second, with inference cost only 1/8 of the flagship Pro model. Enhanced instruction following and batch structured extraction stability plus prompt context caching drastically cut computing overhead for large-scale deployments, balancing ultra-low latency, reliable task accuracy and outstanding cost efficiency. Key Capabilities ● Adaptive Multi-Level Dynamic Reasoning: Features automatic reasoning intensity adjustment, achieving 86.9% accuracy on GPQA Diamond benchmark; delivers near-instant responses for classification & QA while enabling chain-of-thought self-verification for complex business tasks. ● Lightweight High-Frequency Coding Assistance: Excels at scripting, data cleaning and simple API development, ideal for daily R&D auxiliary tasks and automated data pipeline construction. ● Native Batch Four-Modal Structured Extraction: Processes bulk receipts, scanned PDFs, short videos and audio recordings to realize automatic transcription, table parsing and key information standardization for offline asset digitization. ● Million-Token Lossless Long Context: Supports one-time loading of massive documents and full conversation history with stable key detail recall, applicable to knowledge base retrieval and long-turn dialogue memory retention. ● High-Throughput Batch Agent Orchestration: Enables parallel multi-tool invocation, sandbox code execution and web grounding with parameter validation & auto-retry to support large-scale lightweight workflow clusters. ● Prompt Context Caching Mechanism: Reuses cached prompts and dialogue history to cut token consumption by over 40% for repetitive batch workloads, greatly lowering AI adoption barriers for small and medium enterprises. ● Multilingual Enterprise Compliance Processing: Natively supports hundreds of languages with layered content risk control, data desensitization and hallucination mitigation for bulk translation, content moderation and cross-border content production. Technical Strengths 表格 Feature Benefit Ultra-Optimized Sparse MoE Architecture Activates minimal required computing modules to stabilize high-concurrency throughput with tiny latency fluctuation, guaranteeing reliable online service for millions of API requests 2.5× Accelerated First-Token Response Shortens user waiting time, reduces bounce rate for real-time chat and customer service applications and significantly improves end-user interactive experience Lightweight Unified Multimodal Encoder Eliminates redundant multimedia feature data to accelerate batch preprocessing, lowering bandwidth and server resource expenditure for enterprise deployment Context Prompt Caching Avoids repeated full prompt transmission for recurring business tasks, delivering substantial cost savings for large-scale batch industrial workloads Hard-Constrained Structured Output Native JSON Schema validation prevents parsing failures in downstream data pipelines, minimizing manual data correction and improving production stability Specialized Instruction Tuning Delivers consistent outputs for classification, summarization, translation and information extraction tasks, reducing human post-processing workload for bulk business pipelines Capability Ratings 表格 Dimension Rating Notes Reasoning Strong Adaptive reasoning ensures reliable accuracy for standardized business and scientific tasks, slightly inferior to flagship Flash models on high-order complex deduction Coding Moderate Capable of scripting and lightweight business code development, perfectly suited for data automation and daily developer auxiliary scenarios Creative Writing Strong Generates standardized, high-speed commercial, educational and marketing copy at scale for mass content production pipelines Multimodal Strong Retains complete four-modal capability with outstanding performance on bulk receipt, table and scanned document parsing for cost-sensitive multimedia workflows Response Speed Very Fast The highest throughput and lowest latency model in Gemini 3.1 series, optimized for real-time interactive and high-concurrency production environments Context Window Huge Million-token input capacity enables global analysis of document clusters, knowledge bases and extended dialogue histories for enterprise data governance Use Cases ● Real-Time Omnichannel Customer Service: Handles massive concurrent user consultations and intent classification to realize intelligent cost reduction for after-sales service systems. ● Bulk Multilingual Content Operations: Automates translation, content moderation and compliance inspection for cross-border e-commerce copy, short-video subtitles and news materials. ● Financial Receipt Intelligent Digitization: Extracts structured data from invoices, expense forms and delivery slips to automate financial entry and reimbursement workflows. ● Enterprise Mass Document Governance: Executes batch summarization, tagging and classification for policies, manuals and internal archives to streamline unstructured data asset management. ● Lightweight Enterprise Agent Cluster Deployment: Automates data statistics, report generation and work order classification to boost operational efficiency across business departments at scale. ● Consumer-Facing High-Frequency AI Tools: Powers mini-program chatbots, copywriting assistants and educational tutoring applications with controllable computing costs and smooth real-time interaction. ● Enterprise Meeting Asset Archiving: Automatically transcribes, summarizes and archives meeting audios and recordings to build searchable internal corporate knowledge assets.