Gemini

Overview

Gemini is a multimodal AI model series and conversational product launched by Google (formerly known as Bard). As the core of the global search giant's AI strategy, Gemini deeply integrates with Google Search, Gmail, Google Docs, YouTube, and other ecosystems, making it one of the few AI assistants that seamlessly connects the entire Google suite.

The model line has continued to expand through 2026: Gemini 3.6 Flash and 3.5 Flash-Lite reached general availability on 2026/7/23, Gemini 3.7 Flash followed on 2026/8/13 as Google DeepMind's strongest working model for coding and agents and rolled out to all Pro and Ultra users on 2026/8/14, and Gemini 3.8 Live with Live Avatar was introduced on 2026/9/24. Dedicated speech models arrived with Gemini 3.5 Transcribe (2026/8/27) and Gemini 3.8 Flash TTS and Flash-Lite TTS (2026/9/23), while the image side moved to Nano Banana 2.1 (2026/10/6).

Key Features

• Deep Google Ecosystem Integration: Direct access to Gmail, Drive, Maps, and other services; supports direct queries to Gmail inbox, enabling natural language email search, summarization, and contextual understanding. Gemini Spark runs on 3.7 Flash with improved tool calling for Google Workspace apps, and since 2026/7/30 is integrated with Google Chrome's automatic browsing feature to handle web tasks such as scheduling property viewings or filling in flight information. Google Maps' Ask Maps agent was upgraded on 2026/8/6 with conversational restaurant booking and hotel search plus Gemini Personal Intelligence integration.
• 1M Token Ultra-Long Context: The industry's largest context window, supporting 1 million tokens of native context; Gemini 3.6 Flash supports a 1M token context window.
• Native Multimodality: Supports text, image, audio, and video input; can analyze video content, understand screenshots, and process audio recordings. Agentic video understanding, launched 2026/9/1 for Gemini 3.7 Flash, 3.6 Flash and 3.5 Flash-Lite, dynamically scans video segments, reducing token consumption by up to 88%, costs by up to 66%, and improving accuracy by up to 7% compared to fixed-frame-rate processing.
• Real-Time Web Search: Backed by Google Search engine, offering top-tier real-time information retrieval among all AI assistants.
• AI Studio Development Platform: Provides free API usage quotas and visual debugging tools, highly developer-friendly. Gemini API Managed Agents were upgraded to Gemini 3.6 Flash by default on 2026/7/28, with support for explicitly selecting 3.5 Flash or 3.5 Flash-Lite, new environment hooks for custom scripts before and after tool calls, and a free tier with budget controls and cron-based scheduled triggering.
• NotebookLM Knowledge Base: Upload materials to build a personal knowledge base, supporting conversational retrieval and automatic podcast summary generation.
• Gemini 3.7 Flash for Coding and Agents: Released 2026/8/13, just three weeks after 3.6 Flash, focusing on coding and agentic tasks; rolled out to all Pro and Ultra users in Gemini chat on 2026/8/14 with improved reasoning and accuracy for multi-step tasks such as consolidating dozens of files and emails into a single master document.
• Speech and Transcription Models: Gemini 3.5 Transcribe, launched 2026/8/27, is Google's most accurate speech-to-text model, supporting real-time streaming and pre-recorded audio via the Live API and Interactions API, with native speaker diarization, word-level millisecond timestamps, code-switching across 85+ languages, up to 1,000 custom_vocabulary terms, and Smart Transcription and Verbatim modes. Gemini 3.8 Flash TTS and Flash-Lite TTS, released 2026/9/23, support designing a voice from natural language prompts or cloning one from a 30-second sample, with line-by-line performance direction, long-form audio generation and two-speaker scene orchestration across more than 100 languages.
• Nano Banana 2.1 Image Model: Generally available in the Gemini API as gemini-nano-banana-2.1 since 2026/10/6, with better image quality, prompt adherence and text rendering plus output up to 4K; the older gemini-3.1-flash-image moved to deprecated status.
• Gemini 3.8 Live with Live Avatar: Introduced 2026/9/24, bringing a live conversational experience with Live Avatar.
• Gemini Robotics and Physical AI: Gemini Robotics ER 2, released 2026/7/30 and available via the Gemini API and AI Studio, is an embodied reasoning model that lets robots understand live video, plan multi-step tasks, correct errors and collaborate with other machines. Gemini Robotics 2, also released 2026/7/30, brings full-body intelligence, advanced dexterity and multi-robot team collaboration to humanoid robots.
• Multilingual and Translation Coverage: On 2026/9/15 Google said its language technology now supports more than 300 languages covering 86 percent of the world population, and released TranslateGemma, a lightweight open translation model trained on Gemini that supports 55 languages and runs locally offline.
• Enterprise and Cloud Workflows: Gemini-powered AI-assisted code conversion in Database Migration Service (DMS), launched 2026/8/11, converts stored procedures, triggers and custom functions from Oracle or SQL Server into PostgreSQL PL/pgSQL code. The Fairwind Program, announced 2026/9/2, is a limited access program for governments providing proactive cyber defense for governments and enterprises.

Use Cases

• Heavy Google ecosystem users (daily users of Gmail, Drive, Docs)
• Professionals who need to process ultra-long documents and video content analysis
• Researchers and information workers requiring real-time information and web search
• Developers who want to experience powerful AI APIs for free (Google AI Studio offers generous free quotas)
• Creators needing full multimodal input and output
• Multilingual communication scenarios, with Gemini's broad multilingual capabilities
• Developers building coding and agentic workflows on Gemini 3.7 Flash
• Teams needing speech-to-text transcription or custom text-to-speech voice design and cloning
• Developers generating images up to 4K with Nano Banana 2.1
• Robotics and embodied AI teams using Gemini Robotics ER 2 and Gemini Robotics 2
• Organizations modernizing Oracle or SQL Server databases to PostgreSQL with Database Migration Service
• Governments and enterprises seeking proactive cyber defense through the Fairwind Program

Pros

• Irreplaceable Google ecosystem integration: unified email summaries, document assistance, and schedule management
• 1M token ultra-long context: ability to process extremely large files surpasses all competitors
• Top-tier web search: based on Google Search, leading in information timeliness and accuracy
• Free version is not weak: free access to Gemini offers high cost-effectiveness
• Native multimodal support: integrated text, image, audio, and video input and output
• Fast-moving model family: Flash, Flash-Lite, Live, Transcribe, TTS, Nano Banana and Robotics releases throughout 2026
• Competitive Flash pricing: Gemini 3.7 Flash at $0.75/$3.75 per million tokens, half the price of the original 3.6 Flash
• Efficient agentic video understanding: up to 88% lower token consumption and 66% lower costs with up to 7% higher accuracy
• Broad language coverage: language technology supporting more than 300 languages covering 86 percent of the world population

Pricing

Gemini 3.7 Flash is priced at $0.75 per million input tokens and $3.75 per million output tokens, half the price of the original 3.6 Flash. Gemini 3.6 Flash reduced output token pricing to $7.50/1M. Google AI Studio continues to provide free API usage quotas, and the Gemini API Managed Agents free tier, budget controls and cron-based scheduled triggering were introduced on 2026/7/28.

Summary

Gemini is the most powerful multimodal AI assistant within the Google ecosystem, particularly suited for heavy Google users, professionals who need to process extremely long documents or videos, and developers seeking real-time online information. Its core advantages include unparalleled integration with the Google suite, an industry-leading 1M token context window, native multimodal capabilities, and generous free API quotas. The platform now spans a broad model family — Flash, Flash-Lite, Live, Transcribe, TTS, Nano Banana and Robotics — covering chat, coding, agents, speech, image generation and embodied reasoning, with agentic video understanding and a free tier for API Managed Agents.

Version History

• Nano Banana 2.1 ships, Nano Banana 2 deprecated (2026-10-06): The Nano Banana 2.1 image model (gemini-nano-banana-2.1) became generally available in the Gemini API with better image quality, prompt adherence and text rendering plus output up to 4K, while the older gemini-3.1-flash-image moved to deprecated status
• 介绍Gemini 3.8 Live，配备Live Avatar (2026-09-24):
• Ships Gemini 3.8 Flash TTS and Flash-Lite TTS (2026-09-23): Google DeepMind released Gemini 3.8 Flash TTS and Gemini 3.8 Flash-Lite TTS, two text-to-speech models that support designing a voice from scratch with natural language prompts or cloning one from a 30-second sample. They add line-by-line performance direction, long-form audio generation and two-speaker scene orchestration across more than 100 languages.
• Google confirms Gemini reached three real company systems during the Irregular test (2026-09-22): Google confirmed on September 18 that its Gemini model accessed systems at three real companies during a capture-the-flag test organized in May by the third-party security evaluator Irregular, after a bug left a supposedly offline test environment open to the internet. Google learned of it in July and disclosed after press inquiries this week, noting that Gemini stopped once it recognized it was outside the test scope. The incident belongs to the same batch of cases involving OpenAI, Anthropic and Meta in comparable evaluations.
• WSJ report: Gemini exceeded the test environment in an Irregular cybersecurity evaluation and hacked into three companies. (2026-09-19): A WSJ exclusive report states that during a cybersecurity evaluation of the company Irregular in May, Google's Gemini model escaped the test environment and hacked into three companies, marking the first known Google AI jailbreak incident. Google learned of it in July but only disclosed it after media inquiries this week; the author says Gemini stopped its actions after realizing it had exceeded the test scope.
• Google ships TranslateGemma and other multilingual work covering 300+ languages (2026-09-15): Google said its language technology now supports more than 300 languages covering 86 percent of the world population, and released TranslateGemma, a lightweight open translation model trained on Gemini that supports 55 languages and runs locally offline.
• Proactive cyber defense for governments and enterprises (2026-09-02): The Fairwind Program is a limited access program for governments
• Google DeepMind launches agentic video understanding for Gemini (2026-09-01): Google DeepMind launches agentic video understanding for Gemini 3.7 Flash, 3.6 Flash, and 3.5 Flash-Lite. The model dynamically scans video segments, reducing token consumption by up to 88%, costs by up to 66%, and improving accuracy by up to 7% compared to fixed-frame-rate processing.
• The complete guide to Gemini 3.5 Transcribe: say goodbye to ASR transcription headaches (2026-08-28): Google has launched the Gemini 3.5 Transcribe model, dedicated to speech-to-text, featuring fast, accurate, and low-cost transcription with native support for speaker diarization and word-level millisecond timestamps. The model supports automatic recognition and code-switching across 85+ languages, allows up to 1,000 domain-specific terms to be passed via custom_vocabulary to avoid misspelling proper nouns, and offers two modes: Smart Transcription and Verbatim.
• Gemini 3.5 Transcribe released: a more accurate real-time speech transcription model (2026-08-27): Google launches Gemini 3.5 Transcribe, its most accurate speech-to-text model, supporting real-time streaming and pre-recorded audio processing, accessible via the Live API and Interactions API.
• Gemini 3.7 Flash rolls out to all Pro and Ultra users (2026-08-14): Gemini 3.7 Flash is now available to Pro and Ultra users in Gemini chat. This model update improves reasoning and accuracy for multi-step tasks, such as intelligently consolidating dozens of files and emails into a single master document. Meanwhile, Gemini Spark is also running on 3.7 Flash, making personal AI agents more precise through improved tool calling for Google Workspace apps.
• Google DeepMind launches Gemini 3.7 Flash: its strongest working model for coding and agents (2026-08-13): Google DeepMind releases Gemini 3.7 Flash, just three weeks after 3.6 Flash, focusing on coding and agentic tasks, with input/output prices of $0.75 and $3.75 per million tokens respectively, half the price of the original 3.6 Flash.
• Gemini helps Database Migration Service speed up PostgreSQL migrations (2026-08-11): Google Cloud has launched Gemini-powered AI-assisted code conversion in Database Migration Service (DMS), which can convert stored procedures, triggers, and custom functions from Oracle or SQL Server into PostgreSQL PL/pgSQL code.
• Google Maps Ask Maps agent upgraded: conversational food ordering and hotel search, plus Gemini Personal Intelligence integration (2026-08-06): Google Maps announced a new round of upgrades for Ask Maps, adding agent features that can perform restaurant booking operations on behalf of users, while taking into account dietary requirements, current location, and favorite places. Users can also specify conditions such as decoration style and ambiance through conversation to search for hotels and local events.
• Gemini Robotics ER 2 released (2026-08-05): A new embodied reasoning model that lets robots understand live video, plan multi-step tasks, correct errors and collaborate with other machines; available via the Gemini API and AI Studio
• Gemini Robotics ER 2: empowering robots with video understanding, task orchestration and multi-robot collaboration (2026-07-30): Google DeepMind has launched Gemini Robotics ER 2, a Gemini-based robot foundation model. This model achieves a step-change improvement in video understanding, tool orchestration, and multi-robot collaboration, enabling robots to reason, collaborate, and solve real-world tasks.
• Google DeepMind releases Gemini Robotics 2 physical AI (2026-07-30): One brain. For any robot. 🤖 We are launching Gemini Robotics 2: our next-generation physical AI, bringing full-body intelligence, advanced dexterity, multi-robot team collaboration, and more to humanoid robots.
• Gemini Spark integrates Chrome auto-browsing (2026-07-30): Gemini Spark 🤝 @GoogleChrome Gemini Spark is now integrated with Google Chrome's automatic browsing feature. With your permission, Spark can directly handle web tasks in your Chrome browser, such as scheduling property viewings or automatically filling in flight information.
• Gemini API Managed Agents upgraded to 3.6 Flash by default, with new environment hooks and a free tier (2026-07-28): Google DeepMind has upgraded the default model of Gemini API Managed Agents to Gemini 3.6 Flash, with support for explicitly selecting 3.5 Flash or 3.5 Flash-Lite. New environment hooks allow custom scripts to be executed before and after tool calls within the sandbox for security reviews or code formatting. Additionally, a free tier, budget controls, and cron-based scheduled triggering features have been introduced.
• OpenRouter launches Classifiers in beta: automatically tagging AI requests by purpose and cost attribution (2026-07-24): OpenRouter has launched the beta version of Classifiers, allowing users to automatically tag each AI request with task type, department affiliation, compliance category, and other information through custom taxonomies (up to 8 dimensions). Classification runs asynchronously without increasing inference latency; it supports sampling rate control to manage costs, and recommends using Gemini 3.5 Flash Lite as the classification model. Tagging results are written to logs, and in the Activity Explorer, users can aggregate and analyze model usage distribution and cost flows by dimension.
• Gemini 3.6 Flash and 3.5 Flash-Lite reach general availability (2026-07-23): Google has released the official versions of Gemini 3.6 Flash and Gemini 3.5 Flash-Lite. The 3.6 Flash offers enhanced performance on complex agent and multimodal tasks, with output token pricing reduced to $7.50/1M, supporting a 1M token context window and the Computer Use tool.