Google Gemini release history.

202610 RELEASES

1. 10 SEP 2026PRODUCT

   Gemini app for Windows

   • LAUNCHED
   • MONTHLY FLASH

   Google launched a native Gemini app for Windows 10 and 11, available globally. It opens with Alt + Space over any window and connects to Gmail and Drive. Some features, such as Gemini Spark and Omni video generation, need a Google AI subscription.

   Why it mattered. After the Mac app in April, it put Gemini on the desktop for most PC users, making it a direct alternative to other desktop AI assistants.

   AVAILABLE TO

   Free · Google AI Plus · Google AI Pro · Google AI Ultra

   SOURCES

   The Gemini app is now available for Windows ↗Gemini Apps release notes (2026.09.10) ↗

2. 2 SEP 2026MODEL

   Gemini 3.8 Flash

   • GENERALLY AVAILABLE
   • CURRENT
   • MONTHLY FLASH

   Google released Gemini 3.8 Flash as generally available, its most capable Flash model, aimed at long-horizon software engineering and agents. It is more accurate than 3.7 Flash but uses more tokens, at the same introductory price of US$0.75/US$3.75 per million tokens. Gemini 3.8 Flash Cyber went to vetted defenders through the new Fairwind Program.

   Why it mattered. It became Google's recommended default model for new API projects and the Flash model for Pro and Ultra subscribers in the Gemini app, AI Mode and Sheets. Google says it approaches much more expensive frontier models on some agentic coding tests.

   AVAILABLE TO

   Google AI Pro · Google AI Ultra · API · Vertex AI · Gemini Enterprise

   SOURCES

   Introducing Gemini 3.8 Flash and 3.8 Flash Cyber ↗Gemini API release notes (September 2, 2026) ↗

   BENCHMARKS

   Gemini 3.8 Flash →

3. 13 AUG 2026MODEL

   Gemini 3.7 Flash

   • GENERALLY AVAILABLE
   • SUPERSEDED
   • MONTHLY FLASH

   Google released Gemini 3.7 Flash as generally available, three weeks after 3.6 Flash, with gains in coding, web development and agentic work (43.6% on FrontierCode against 34.4% for 3.6 Flash). It has an introductory API price of US$0.75/US$3.75 per million input/output tokens until 31 December 2026.

   Why it mattered. It halved the introductory price of the Flash tier while lifting coding quality. It also became the model behind Gemini Spark for Google AI Pro and Ultra subscribers.

   AVAILABLE TO

   Google AI Pro · Google AI Ultra · API · Vertex AI · Gemini Enterprise

   SOURCES

   Introducing Gemini 3.7 Flash ↗Gemini API release notes (August 13, 2026) ↗

   BENCHMARKS

   Gemini 3.7 Flash →

4. 21 JUL 2026MODEL

   Gemini 3.6 Flash and 3.5 Flash-Lite

   • GENERALLY AVAILABLE
   • CURRENT
   • MONTHLY FLASH

   Google released Gemini 3.6 Flash and Gemini 3.5 Flash-Lite as generally available. 3.6 Flash uses fewer output tokens than 3.5 Flash and costs US$1.50/US$7.50 per million input/output tokens, and 3.5 Flash-Lite is the cheapest Flash-class model at US$0.30/US$2.50. Google also announced Gemini 3.5 Flash Cyber, a vulnerability-finding model limited to governments and trusted partners through CodeMender.

   Why it mattered. 3.6 Flash replaced 3.5 Flash as the everyday model for all Gemini app users at a lower price, and 3.5 Flash-Lite became the recommended low-cost model for new API projects. Google said Gemini 3.5 Pro was still testing with partners and that pre-training for Gemini 4 had begun.

   AVAILABLE TO

   Free · Google AI Pro · Google AI Ultra · API · Vertex AI · Gemini Enterprise

   SOURCES

   Introducing Gemini 3.6 Flash, 3.5 Flash-Lite, and 3.5 Flash Cyber ↗Gemini API release notes (July 21, 2026) ↗Gemini Apps release notes (2026.07.21) ↗

   BENCHMARKS

   Gemini 3.6 Flash →Gemini 3.5 Flash-Lite →

5. 19 MAY 2026AGENT

   Gemini Spark

   • TRUSTED TESTING
   • GEMINI 3 AND OMNI

   Google introduces Gemini Spark, a personal AI agent powered by 3.5 Flash that can help navigate digital life and act under user direction.

   Why it mattered. Spark made the Gemini app's agentic direction concrete as an always-available personal agent.

   AVAILABLE TO

   Trusted testers · Google AI Ultra

   SOURCE

   Gemini 3.5 ↗

6. 19 MAY 2026MODEL

   Gemini Omni

   • WATERSHED
   • LAUNCHED
   • CURRENT
   • GEMINI 3 AND OMNI

   Gemini Omni Flash is introduced for Google Flow and Flow Music, combining Gemini intelligence with generative media models for video creation and editing.

   Why it mattered. Omni extended Gemini from assistant and developer intelligence into any-input, any-output creative generation.

   AVAILABLE TO

   Google AI subscribers · Creators · Flow users

   SOURCE

   Flow and Gemini Omni ↗

7. 19 MAY 2026PRODUCT

   Agentic Gemini app

   • ANNOUNCED
   • GEMINI 3 AND OMNI

   Google announces a more agentic Gemini app with Daily Brief, proactive help, tailored response designs and broader monthly usage.

   Why it mattered. The app moved toward a proactive assistant that organises information and experiences rather than waiting for prompts.

   AVAILABLE TO

   Gemini app

   SOURCE

   Next evolution of the Gemini app ↗

8. 19 MAY 2026MODEL

   Gemini 3.5 Flash

   • GA
   • SUPERSEDED
   • GEMINI 3 AND OMNI

   Gemini 3.5 launches with 3.5 Flash, a generally available model built for complex agentic workflows, coding and multimodal tasks.

   Why it mattered. Gemini 3.5 positioned Flash as a frontier-level agent model rather than just a fast smaller model.

   AVAILABLE TO

   Everyone · Developers · Enterprises

   SOURCE

   Gemini 3.5 ↗

   BENCHMARKS

   Gemini 3.5 Flash →

9. 19 FEB 2026MODEL

   Gemini 3.1 Pro

   • PREVIEW
   • CURRENT
   • GEMINI 3 AND OMNI

   Google released Gemini 3.1 Pro, an upgraded Pro model for complex reasoning, scoring a verified 77.1% on ARC-AGI-2, more than double Gemini 3 Pro. It rolled out in the Gemini app (higher limits on Pro and Ultra) and NotebookLM, and in preview for developers and enterprises.

   Why it mattered. It became the Pro model in the Gemini app and the recommended replacement for Gemini 3 Pro Preview, which was shut down on 9 March 2026. It remained Google's top Pro model through September 2026 because Gemini 3.5 Pro had not shipped.

   AVAILABLE TO

   Free · Google AI Pro · Google AI Ultra · API · Vertex AI · Gemini Enterprise

   SOURCES

   Gemini 3.1 Pro: A smarter model for your most complex tasks ↗Gemini Apps release notes (2026.02.19) ↗Gemini API release notes (February 19, 2026) ↗

   BENCHMARKS

   Gemini 3.1 Pro →

10. 12 FEB 2026MODEL

    Gemini 3 Deep Think

    • LAUNCHED
    • CURRENT
    • GEMINI 3 AND OMNI

    Gemini 3 Deep Think launches as a major upgrade for science, research and engineering challenges, with access for Google AI Ultra subscribers.

    Why it mattered. Deep Think extended Gemini 3's frontier into high-effort research reasoning and specialised problem solving.

    AVAILABLE TO

    Google AI Ultra · Researchers · Enterprises

    SOURCE

    Gemini 3 Deep Think ↗

202513 RELEASES

1. 17 DEC 2025MODEL

   Gemini 3 Flash

   • LAUNCHED
   • SUPERSEDED
   • GEMINI 3 AND OMNI

   Gemini 3 Flash launches as a frontier-intelligence model built for speed and becomes the default model in the Gemini app.

   Why it mattered. Flash brought Gemini 3 capability to faster, broader everyday usage.

   AVAILABLE TO

   Gemini app · AI Mode · Developers

   SOURCE

   Gemini 3 Flash ↗

2. 20 NOV 2025MODEL

   Nano Banana Pro (Gemini 3 Pro Image)

   • LAUNCHED
   • SUPERSEDED
   • GEMINI 3 AND OMNI

   Google launched Nano Banana Pro, built on Gemini 3 Pro, for image generation and editing with legible text in multiple languages, up to 14 input images and output up to 4K. It rolled out globally in the Gemini app (limited for free users, higher quotas on AI Plus, Pro and Ultra), in the Gemini API and Google AI Studio, and in Workspace and Google Ads.

   Why it mattered. It made studio-quality image editing and graphics with accurate text usable inside Gemini, and every image carries a SynthID watermark that Gemini can now check.

   AVAILABLE TO

   Free · AI Plus · AI Pro · AI Ultra · API

   SOURCE

   Introducing Nano Banana Pro ↗

3. 18 NOV 2025AGENT

   Google Antigravity

   • LAUNCHED
   • GEMINI 3 AND OMNI

   Google releases Antigravity, an agentic development platform where Gemini agents can plan, code, use the terminal and validate work in a browser.

   Why it mattered. Antigravity made agentic software development a first-class Gemini platform surface.

   AVAILABLE TO

   Developers

   SOURCE

   A new era with Gemini 3 ↗

4. 18 NOV 2025AGENT

   Gemini Agent

   • EXPERIMENTAL
   • GEMINI 3 AND OMNI

   Google AI Ultra subscribers can try Gemini Agent in the Gemini app for complex workflows like organising Gmail or planning services.

   Why it mattered. The Gemini app moved from conversation into supervised multi-step action.

   AVAILABLE TO

   Google AI Ultra

   SOURCE

   Gemini 3 app updates ↗

5. 18 NOV 2025MODEL

   Gemini 3 Pro preview

   • WATERSHED
   • PREVIEW
   • SUPERSEDED
   • GEMINI 3 AND OMNI

   Gemini 3 begins with Gemini 3 Pro in preview across the Gemini app, AI Studio, Vertex AI, Search AI Mode and Antigravity.

   Why it mattered. Gemini 3 unified multimodal understanding, coding, reasoning, planning and agentic workflows into Google's next frontier model family.

   AVAILABLE TO

   Gemini app · Developers · Vertex AI · Search

   SOURCE

   A new era with Gemini 3 ↗

6. 26 AUG 2025PRODUCT

   Nano Banana image editing

   • WATERSHED
   • LAUNCHED
   • SUPERSEDED
   • GEMINI 3 AND OMNI

   Google introduces an updated Gemini image editing model from Google DeepMind focused on likeness consistency, blending, multi-turn edits and style transfer.

   Why it mattered. Nano Banana made Gemini a viral creative tool and showcased native multimodal generation in a consumer surface.

   AVAILABLE TO

   Gemini app · Creators

   SOURCE

   Image editing in Gemini ↗

7. 1 AUG 2025MODEL

   2.5 Deep Think

   • ROLLING OUT
   • SUPERSEDED
   • GEMINI 3 AND OMNI

   Google rolls out Deep Think in the Gemini app for Google AI Ultra subscribers and shares access with select mathematicians.

   Why it mattered. Deep Think gave Gemini a visible high-effort reasoning mode for complex math, code and scientific tasks.

   AVAILABLE TO

   Google AI Ultra · Researchers

   SOURCE

   Try Deep Think ↗

8. 17 JUN 2025MODEL

   2.5 Pro and Flash GA

   • GA
   • AGENTIC 2.X

   Gemini 2.5 Pro and Flash are released as stable, generally available models for production applications.

   Why it mattered. This made the 2.5 generation reliable enough for scaled developer and enterprise use.

   AVAILABLE TO

   Developers · Vertex AI · Enterprises

   SOURCE

   2.5 family expands ↗

9. 20 MAY 2025MODEL

   Gemini 2.5 I/O updates

   • ANNOUNCED
   • AGENTIC 2.X

   At I/O 2025, Google announces 2.5 Flash improvements, Deep Think for 2.5 Pro, native audio output, advanced safeguards and more developer controls.

   Why it mattered. The 2.5 family became a full platform with reasoning, speed, audio, safety and developer affordances.

   AVAILABLE TO

   Developers · Gemini app · Enterprises

   SOURCE

   Gemini 2.5 I/O updates ↗

10. 17 APR 2025MODEL

    Gemini 2.5 Flash preview

    • PREVIEW
    • SUPERSEDED
    • AGENTIC 2.X

    Google released an early preview of Gemini 2.5 Flash in the Gemini API, Google AI Studio, Vertex AI and the Gemini app. It was Google's first fully hybrid reasoning model: developers can turn thinking on or off and set a thinking budget of 0 to 24,576 tokens.

    Why it mattered. It brought reasoning to Google's fast, low-cost model tier and let developers trade off quality against cost and speed per request.

    AVAILABLE TO

    Gemini app · API

    SOURCE

    Start building with Gemini 2.5 Flash ↗

11. 25 MAR 2025MODEL

    Gemini 2.5 Pro Experimental

    • WATERSHED
    • EXPERIMENTAL
    • SUPERSEDED
    • AGENTIC 2.X

    Gemini 2.5 Pro Experimental debuts as Google's most intelligent model, with reasoning built into the model family.

    Why it mattered. Gemini moved from optional thinking modes toward a generation designed around reasoning by default.

    AVAILABLE TO

    Gemini Advanced · AI Studio

    SOURCE

    Gemini 2.5 ↗

12. 12 MAR 2025AGENT

    Gemini Robotics

    • INTRODUCED
    • AGENTIC 2.X

    Google DeepMind introduces Gemini Robotics and Gemini Robotics-ER, bringing Gemini 2.0's multimodal reasoning into robotics.

    Why it mattered. Robotics moved Gemini's agentic ambitions from screens and browsers into embodied systems.

    AVAILABLE TO

    Roboticists · Researchers

    SOURCE

    Gemini Robotics ↗

13. 5 FEB 2025MODEL

    Gemini 2.0 Flash GA

    • GA
    • SUPERSEDED
    • AGENTIC 2.X

    Google makes Gemini 2.0 Flash generally available through the Gemini API in Google AI Studio and Vertex AI.

    Why it mattered. This moved Gemini 2.0 from experimental excitement into production deployment.

    AVAILABLE TO

    Developers · Vertex AI · AI Studio

    SOURCE

    Gemini 2.0 is now available ↗

202414 RELEASES

1. 19 DEC 2024MODEL

   Gemini 2.0 Flash Thinking Experimental

   • EXPERIMENTAL
   • SUPERSEDED
   • AGENTIC 2.X

   Google released Gemini 2.0 Flash Thinking Mode as an experimental public preview in the Gemini API and Google AI Studio. It is a test-time compute model that shows its thought process while it generates a response.

   Why it mattered. It started Google's line of 'thinking' models that reason before answering, and it showed users the reasoning behind each answer.

   AVAILABLE TO

   API

   SOURCE

   Gemini API release notes (changelog) ↗

2. 11 DEC 2024AGENT

   Project Mariner

   • TRUSTED TESTING
   • AGENTIC 2.X

   Project Mariner explores using Gemini to reason across browser screens and web elements, then complete tasks through a Chrome extension.

   Why it mattered. Mariner put computer-use agents into the Gemini story and forced safety questions around web actions and user confirmation.

   AVAILABLE TO

   Trusted testers

   SOURCE

   Introducing Gemini 2.0 ↗

3. 11 DEC 2024AGENT

   Jules code agent

   • EXPERIMENTAL
   • AGENTIC 2.X

   Jules is introduced as an experimental AI-powered code agent that can plan and execute coding tasks under developer supervision.

   Why it mattered. It put Gemini into the software engineering agent lane before the later Antigravity platform.

   AVAILABLE TO

   Developers

   SOURCE

   Introducing Gemini 2.0 ↗

4. 11 DEC 2024MODEL

   Gemini 2.0 Flash

   • WATERSHED
   • EXPERIMENTAL
   • SUPERSEDED
   • AGENTIC 2.X

   Google introduces Gemini 2.0 and releases an experimental Gemini 2.0 Flash with stronger performance, native tool use and multimodal output capabilities.

   Why it mattered. Gemini 2.0 reframed the model family around agents, action, native tool use and richer input-output loops.

   AVAILABLE TO

   Developers · Gemini users · Trusted testers

   SOURCE

   Introducing Gemini 2.0 ↗

5. 11 DEC 2024AGENT

   Deep Research

   • LAUNCHED
   • AGENTIC 2.X

   Google launches Deep Research in Gemini Advanced to browse, reason and compile multi-step research reports.

   Why it mattered. Deep Research turned Gemini into a task-oriented research workflow rather than a single-answer chatbot.

   AVAILABLE TO

   Gemini Advanced

   SOURCE

   Introducing Gemini 2.0 ↗

6. 14 NOV 2024PRODUCT

   Gemini app for iPhone

   • LAUNCHED
   • LONG CONTEXT MOBILE

   The Gemini app becomes available on iPhone, including Gemini Live, image generation with Imagen 3 and app connections.

   Why it mattered. A standalone iPhone app gave Gemini a cleaner cross-platform assistant presence beyond Android and the web.

   AVAILABLE TO

   iPhone · iOS

   SOURCE

   Gemini app for iPhone ↗

7. 13 AUG 2024AGENT

   Gemini Live rolls out

   • ROLLING OUT
   • LONG CONTEXT MOBILE

   Gemini Live begins rolling out in English to Gemini Advanced subscribers on Android with hands-free conversations and 10 voices.

   Why it mattered. Live made Gemini feel more like a conversational assistant than a prompt box.

   AVAILABLE TO

   Gemini Advanced · Android

   SOURCE

   Gemini makes your mobile device powerful ↗

8. 14 MAY 2024AGENT

   Project Astra

   • RESEARCH PREVIEW
   • LONG CONTEXT MOBILE

   Project Astra demonstrates real-time multimodal understanding for an assistant that can see, hear, remember and respond naturally.

   Why it mattered. Astra became the north-star demo for Gemini as a live, ambient assistant rather than a text-only chatbot.

   AVAILABLE TO

   Trusted testers · Researchers

   SOURCE

   Gemini breaks new ground ↗

9. 14 MAY 2024MODEL

   2M context preview

   • WAITLIST
   • LONG CONTEXT MOBILE

   Google announces a 2 million-token context window for Gemini 1.5 Pro via waitlist in AI Studio and Vertex AI.

   Why it mattered. The 2M preview pushed long-context competition into a new tier for codebases, video, audio and document analysis.

   AVAILABLE TO

   Developers · Cloud customers

   SOURCE

   Gemini breaks new ground ↗

10. 14 MAY 2024MODEL

    Gemini 1.5 Flash

    • PREVIEW
    • SUPERSEDED
    • LONG CONTEXT MOBILE

    Google introduces Gemini 1.5 Flash as a lighter model optimised for speed, high-volume tasks and multimodal reasoning with long context.

    Why it mattered. Flash established the high-throughput model lane that later became central to Gemini app defaults and developer workloads.

    AVAILABLE TO

    Developers · Vertex AI · AI Studio

    SOURCE

    Gemini breaks new ground ↗

11. 15 FEB 2024MODEL

    Gemini 1.5 Pro

    • WATERSHED
    • PREVIEW
    • SUPERSEDED
    • GEMINI 1 LAUNCH

    Gemini 1.5 Pro debuts with MoE architecture, strong multimodal performance and up to a 1 million-token context window in private preview.

    Why it mattered. Long context became Gemini's clearest technical differentiator and opened new workflows across code, video, audio and documents.

    AVAILABLE TO

    Developers · Vertex AI · AI Studio

    SOURCE

    Our next-generation model: Gemini 1.5 ↗

12. 8 FEB 2024PLAN

    Duet AI becomes Gemini for Workspace

    • ANNOUNCED
    • GEMINI 1 LAUNCH

    Google says AI Premium subscribers will soon be able to use Gemini in Gmail, Docs, Slides, Sheets and more, formerly Duet AI.

    Why it mattered. Workspace became a major distribution path for Gemini as an enterprise productivity layer.

    AVAILABLE TO

    Workspace · AI Premium

    SOURCE

    Bard becomes Gemini ↗

13. 8 FEB 2024PLAN

    Gemini Advanced with Ultra 1.0

    • LAUNCHED
    • GEMINI 1 LAUNCH

    Gemini Advanced launches with access to Ultra 1.0 through the Google One AI Premium plan.

    Why it mattered. Advanced created Gemini's premium consumer tier and a channel for Google's largest model.

    AVAILABLE TO

    Gemini Advanced · Google One AI Premium

    SOURCE

    Bard becomes Gemini ↗

14. 8 FEB 2024PRODUCT

    Bard becomes Gemini

    • WATERSHED
    • LAUNCHED
    • GEMINI 1 LAUNCH

    Google rebrands Bard as Gemini, making the product name match the model family and repositioning the app as a Google AI assistant.

    Why it mattered. The rebrand unified the consumer assistant around Gemini and retired the Bard identity.

    AVAILABLE TO

    Gemini users

    SOURCE

    Bard becomes Gemini ↗

20236 RELEASES

1. 13 DEC 2023PLATFORM

   Gemini API

   • LAUNCHED
   • GEMINI 1 LAUNCH

   Google makes Gemini Pro available to developers and enterprises through the Gemini API in Google AI Studio and Vertex AI.

   Why it mattered. The API turned Gemini from a product model into an ecosystem model for builders and Cloud customers.

   AVAILABLE TO

   Developers · Vertex AI · AI Studio

   SOURCE

   Introducing Gemini ↗

2. 6 DEC 2023MODEL

   Gemini 1.0

   • WATERSHED
   • LAUNCHED
   • SUPERSEDED
   • GEMINI 1 LAUNCH

   Google launches Gemini 1.0 in three sizes, positioning it as the company's most capable general model for text, image, audio, video and code.

   Why it mattered. Gemini 1.0 marked the start of Google's named frontier model family and unified research, product, device and developer strategy.

   AVAILABLE TO

   Bard · Developers · Pixel · Cloud

   SOURCE

   Introducing Gemini ↗

3. 19 SEP 2023AGENT

   Bard Extensions

   • LAUNCHED
   • BARD FOUNDATION

   Google launches Bard Extensions in English so Bard can retrieve and synthesize information from Google apps and services in one conversation.

   Why it mattered. Extensions gave Bard a uniquely Google path toward personal, app-connected assistance before the Gemini rebrand.

   AVAILABLE TO

   Bard users · Workspace users

   SOURCE

   Bard can now connect to your Google apps ↗

4. 10 MAY 2023MODEL

   PaLM 2 powers Bard

   • LAUNCHED
   • SUPERSEDED
   • BARD FOUNDATION

   Google introduces PaLM 2 with multilingual, reasoning, and coding gains, and says PaLM 2 is powering major improvements in Bard.

   Why it mattered. PaLM 2 gave Bard its first broad capability jump before the Gemini model family was ready.

   AVAILABLE TO

   Bard users · Developers · Workspace

   SOURCE

   Introducing PaLM 2 ↗

5. 21 MAR 2023PRODUCT

   Bard public preview

   • LAUNCHED
   • BARD FOUNDATION

   Google starts opening access to Bard as an early experiment for writing, brainstorming, learning, and search-adjacent exploration.

   Why it mattered. This turned Bard from a teaser into a real product with user feedback loops, safety guardrails, and a path toward broader availability.

   AVAILABLE TO

   U.S. · U.K. · Early users

   SOURCE

   Try Bard and share your feedback ↗

6. 6 FEB 2023PRODUCT

   Bard announced

   • WATERSHED
   • PREVIEW
   • SUPERSEDED
   • BARD FOUNDATION

   Google announces Bard as an experimental conversational AI service powered by LaMDA, opening it first to trusted testers.

   Why it mattered. Bard created the direct consumer surface that later became Gemini, giving Google a place to ship frontier model capability quickly.

   AVAILABLE TO

   Trusted testers

   SOURCE

   An important next step on our AI journey ↗