Google DeepMind’s multimodal AI assistant. Gemini 3.5 Flash became the broad default for the Gemini app and AI Mode in Search at Google I/O on May 19, 2026. Google positions it as the first Gemini 3.5 model, optimized for fast agentic and coding workflows, with Gemini 3.5 Pro still to follow.
The subscription story also changed. Google AI Pro remains the mainstream $19.99/mo tier, while Google AI Ultra now starts at $100/mo and the previous top tier is now $200/mo instead of the old $249.99 price point. Gemini is best understood as a Google bundle: Search, Workspace, Android, Veo/Imagen media tools, Antigravity, AI Studio, and Google One benefits all affect the buying decision.
Recent developments (April-May 2026)
Verified 2026-05-22.
- May 21: Google expanded Gemini into Android agents and smart-home partner infrastructure. ADK for Kotlin, ADK for Android, and Gemini for Home show Gemini spreading into developer orchestration, on-device agents, home APIs, cameras, speakers, and provider bundles.
- May 20: Google Marketing Live turned Gemini into an ads, analytics, creative, YouTube, and commerce layer. Marketers now need to treat Gemini as part of paid acquisition and product-discovery infrastructure, not only a chatbot or Search model.
- May 19: Google I/O made Gemini 3.5 Flash the default AI layer for Search, apps, and subscriptions. Google AI Ultra now starts at $100/mo, the old top tier dropped to $200/mo, and AI Mode in Search now uses Gemini 3.5 Flash globally.
- May 19: Gemini Omni launched as Google’s Gemini-native video creation and editing model for Google AI Plus, Pro, and Ultra subscribers across Gemini, Flow, Flow Music, and YouTube. Treat it as the Google conversational-editing route; keep Veo 3.1 as the cleaner Google API pricing is published.
- May 19: Google Antigravity 2.0 and Managed Agents turned Gemini into a developer-agent stack. Managed Agents in the Gemini API run the Antigravity harness in isolated Linux environments and extend Gemini from chat into agent runtime infrastructure.
- May 19: Gemini 3.5 Flash landed in GitHub Copilot, giving Google’s new model immediate distribution inside a mainstream coding assistant outside Google’s own IDE stack.
- May 12: Google rebranded Android around Gemini Intelligence and launched Googlebook at the Android Show: I/O Edition. Gemini Intelligence reads the screen and completes cross-app tasks; Googlebook is a new premium AI-laptop line shipping fall 2026 with Acer, Asus, Dell, HP, and Lenovo.
- May 11: Google Threat Intelligence Group disclosed it caught the first AI-authored zero-day exploit before mass deployment. The attacker used an AI model to write a 2FA bypass for a popular sysadmin tool; Google has not named the model or the tool.
- May 5: Google released MTP drafters to make Gemma 4 inference up to 3x faster. Gemma is the open-weight sibling to Gemini, and the drafter release improves Google’s local and self-hosted model story.
- May 5: Gemini API File Search adds multimodal RAG, metadata filters, and page citations, tightening Google’s managed retrieval story for agents and assistants that mix documents and visual assets.
- May 2: A reported Gemini UI string points to a possible Omni video-generation surface ahead of Google I/O. Treat Omni as speculative until Google confirms it; the strategic watch item is whether Gemini unifies Veo video and Nano Banana image generation under one media workflow.
- May 1: The Pentagon expanded classified-network AI access to eight major vendors, including Google.
- April 30: Gemini started rolling out to cars with Google built in, replacing the older Google Assistant car experience with a more conversational assistant surface. US English support begins first, with broader availability expected over time.
- April 30: EU preliminary findings could force Google to open Android AI integrations. If final measures are adopted, Google may need to give rivals more access to Android assistant surfaces, potentially narrowing Gemini’s default-assistant advantage.
- April 28: Google Translate added AI pronunciation practice on Android. The update shows Gemini-era language features spreading through high-volume Google products, not only the standalone Gemini app.
- April 27: Pentagon users gained access to Gemini 3.1 Pro through GenAI.mil as Defense One reported more than 1.3 million active users and more than 100,000 agents built on the platform.
- April 27: Google DeepMind announced a South Korea AI-for-science partnership, including an AI Campus in Seoul and work with Korean research institutions on life sciences, energy, weather, and climate.
- April 26: Gemini Embedding 2 reaches GA, strengthening Google’s RAG and semantic-search stack around Gemini.
- April 26: Google AI Studio gets a vibe-coding subscription signal, pointing to a more explicit paid path for app-building workflows.
- April 26: Google expands Deep Research on Max, reinforcing Ultra/Max as the tier for heavier autonomous research workloads.
- April 26: Google follows up on TPU infrastructure for the agentic era, keeping Gemini’s compute-cost story central to its enterprise-agent push.
- April 24: A retrieval-poisoning demo showed how fake web evidence can make AI systems repeat a false claim. Gemini search and Deep Research workflows need independent-source checks for low-coverage claims.
- April 23: Gemini Enterprise Agent Platform launches with Agent Studio, A2A Orchestration, Registry, Identity, Gateway, and Observability. Google’s most complete enterprise-agent stack to date, built on Vertex AI.
- April 23: AI Industry Roundup, April 23 put Gemini’s Cloud Next agent and TPU announcements beside GPT-5.5, Grok, and EU AI funding signals.
- April 23: Agentic Data Cloud ships: Knowledge Catalog, Data Agent Kit, cross-cloud AI-native Lakehouse. Solves the agent-grounding problem that stalled most 2025 agent pilots.
- April 23: 8th-gen TPU 8t (training) and TPU 8i (inference) unveiled on Cloud Next Day 2. TPU 8t delivers 3x Ironwood performance per superpod; TPU 8i triples on-chip SRAM for agent-serving. Staged availability 2026-2027.
- April 23: Google + Wiz ship Agentic Defense: first post-acquisition Wiz integration. Threat detection, detection engineering, and remediation agents on the Gemini Enterprise Agent Platform.
- April 23: $750M partner budget for SIs, consultancies, and startup resellers to distribute Gemini Agents through 2027.
- April 22: Google unveils Ironwood TPU at Cloud Next 2026: 7th-gen, inference-first, 10x peak perf vs TPU v5p, 192 GB HBM3E per chip. Generally available on Google Cloud; expected to power Gemini serving migration through 2026 with downward pricing pressure on Flash and Pro tiers.
- April 22: Google formalizes four-partner silicon coalition (Broadcom, MediaTek, Marvell, TSMC). Roadmap to TPU v8 on TSMC 2nm by late 2027. MediaTek “Zebrafish” inference chip targets 20-30% lower per-token cost than Ironwood.
- April 17: Chrome AI Mode launches desktop side-by-side view. Click a link in AI Mode, Chrome opens the page beside the AI pane instead of replacing it. Plus menu pulls content from open tabs into queries. Distributes Gemini to Chrome’s massive install base as the default browser AI.
- April 21: AI Industry Roundup, April 21 covered Samsung-Gemini expansion, Apple-Gemini Siri reporting, and AI Index follow-ups around Gemini’s consumer distribution.
- April 16: Google in talks to deploy Gemini on Pentagon classified networks. “All lawful purposes” framing similar to OpenAI’s February DoD deal. Google proposed safeguards against domestic surveillance and autonomous weapons use. Reversal of post-Maven 2018 posture.
- April 17: Google Photos ↔ Gemini integration goes live as an opt-in. Paid subscribers can generate images with their own faces and locations via Nano Banana without manually uploading reference photos. First major consumer integration where one toggle unlocks broad library access for generative models.
- April 15: Google Whisk + Doppl shutdown announced for April 30. Doppl’s virtual try-on capability folds into Google Search and Shopping. Whisk fully sunsets. Pattern of Google retiring experimental AI apps into core products continues.
- April 10: Google Gemini ships 3D model generation inside the app. Text or image input produces 3D assets for AR/VR workflows.
- April 2: Google Gemma 4 released under Apache 2.0. Open-weight sibling family to Gemini, fully permissive for commercial use, competes with Llama 4 on open-weight flagship positioning.
System Verdict
Pick Gemini if you live in Google Workspace, rely on Google Search, need image/video tools inside one account, or want the most integrated Google-side assistant bundle. Gemini 3.5 Flash is now the default model in the Gemini app and AI Mode in Search, while AI Pro at $19.99 remains the mainstream paid tier. Gemini also bundles Workspace, Android, Search, Veo/Imagen media tools, Antigravity, and Google One benefits more tightly than ChatGPT or Claude can.
Skip it if your workload is agentic coding, nuanced long-form writing, or anything outside Google’s ecosystem. Claude Opus 4.7 leads on agentic coding (Claude Code CLI), long-form reasoning, and prose voice; ChatGPT leads on Codex, custom-GPT marketplace, and operator-style agent mode; Midjourney still beats Nano Banana 2 on stylized image quality; Perplexity produces tighter citation-first research than Deep Research.
Who pays which tier: Free for casual access, Google AI Plus for light users, Google AI Pro $19.99/mo for most individuals, AI Ultra $100/mo for higher agent/media/developer usage, and AI Ultra $200/mo only when you need the highest limits or top-tier media/agent features. Verify regional rollout and account eligibility before upgrading.
Key Facts
| Current broad default | Gemini 3.5 Flash in Gemini app and AI Mode in Search as of May 19, 2026 |
| Next frontier model | Gemini 3.5 Pro announced as in testing and expected after I/O |
| Context window | Model- and surface-specific; verify current Gemini 3.5 API docs before quoting production limits |
| API pricing | Check current Gemini API pricing for Gemini 3.5; older Gemini 3.1 Pro rates should not be reused without verification |
| Subscription pricing | Free · AI Plus · AI Pro $19.99 · AI Ultra $100 · AI Ultra $200 |
| Image generation | Nano Banana 2 (Pro) · Nano Banana Pro (Ultra) |
| Video generation | Veo 3.1 / Flow surfaces plus Gemini Omni rollout for Google AI Plus, Pro, and Ultra subscribers |
| Workspace integration | Native in Docs, Sheets, Gmail, Drive, Meet, Chrome |
| Deep Research / agents | Deep Research, Daily Brief, Gemini Spark, Search agents, and Managed Agents vary by plan, region, and rollout stage |
| Android | Gemini Intelligence rebrand launched May 12, 2026 with cross-app task completion and on-screen reading; default assistant since March 2026 |
| Hardware | Googlebook AI-laptop line announced May 12, 2026 with Acer, Asus, Dell, HP, Lenovo; ships fall 2026 |
| Storage bundled | Google One storage included by plan; verify current storage allotments on the Google AI subscriptions page |
| Developer / RAG stack | , AI Studio Android support, Gemini Embedding |
| Security | Google Threat Intelligence Group disclosed first AI-authored zero-day exploit thwarted before mass deployment on May 11, 2026 |
Every data point above verified against vendor sources on 2026-05-22. See Sources.
What it actually is
A single consumer + API product spanning text chat, multimodal analysis (images, audio, video, full repos), autonomous Deep Research, native image generation, native video generation through Gemini Omni and Veo 3.1 media surfaces, real-time voice, and in-app workflow integration across Docs, Sheets, Gmail, Drive, Search, Android, Chrome, and Google Cloud. The subscription bundle now matters as much as the model name.
The real moats are ecosystem, multimodal breadth, and vertically integrated infrastructure. Gemini 3.5 Flash now sits inside Search, the Gemini app, Antigravity, and the Gemini API, while Google controls the surrounding Workspace, Android, YouTube, Cloud, TPU, and Google One surfaces. Gemini Embedding 2, Managed Agents, and TPU-backed serving make Google’s case stronger for enterprise RAG and agent workloads where retrieval quality, execution environment, and per-token cost matter together.
For Google-stack users, switching cost to ChatGPT or Claude isn’t about model quality. It’s about giving up native Docs/Sheets/Gmail integration and the storage bundle.
When to pick Gemini
- You live in Google Workspace. Draft in Docs, analyze in Sheets, summarize in Gmail, organize in Drive. All native, no context switching, no copy-paste.
- You want multimodal output in one subscription. Gemini’s image, video, Search, Workspace, Android, and agent surfaces are increasingly bundled through one Google AI account.
- You run autonomous research reports. Deep Research browses dozens of sources and synthesizes into cited reports. Max/Ultra is increasingly the intended tier for heavier research usage; AI Pro is still best for occasional reports.
- You want Google’s agent stack, not just chat. Antigravity 2.0 and Managed Agents in the Gemini API make Gemini relevant for coding-agent and cloud-agent workflows.
- You’re on Android. Gemini became the default assistant in March 2026 across Android Auto and is rolling out to replace Google Assistant on all Android devices.
- You value bundled storage. 5TB at AI Pro replaces standalone Google One. Effectively the AI tier costs ~$10/mo after netting out the storage value.
When to pick something else
- Terminal-native agentic coding: Claude. Claude Code remains the tighter terminal loop; Gemini’s strongest coding story is now Antigravity and Managed Agents.
- Nuanced long-form writing: Claude. Opus 4.7 produces more distinctive prose voice. Gemini defaults to corporate-safe phrasing.
- Broadest plugin / custom-agent ecosystem: ChatGPT. GPT Store has no Gemini equivalent. Gems are smaller-scale.
- Best-quality image generation: Midjourney. Nano Banana 2 is strong on text rendering and personal-context images but still trails Midjourney v7 on stylized photoreal output.
- Citation-first research: Perplexity. Cleaner source management and faster inline citations. Better for ongoing research workflows than Deep Research’s one-shot reports.
- Privacy-sensitive document processing: Claude. Anthropic’s no-training-on-data default at the Team tier is a cleaner posture than Google’s consumer privacy model.
Pricing
Subscription pricing via Google AI subscription page and Google’s May 19, 2026 I/O subscription update:
| Plan | Price | Models | Storage | Who’s it for |
|---|---|---|---|---|
| Free | $0 | Gemini 3.5 Flash access with limits | Varies by Google Account | Casual daily use |
| AI Plus | Regional / current-page pricing | Higher limits and selected Gemini media features | Check current plan page | Light users who need more quota than Free |
| AI Pro | $19.99/mo | Main paid Gemini bundle, Search/Workspace/Gemini app features, selected media and agent features | Check current plan page | Most individuals land here |
| AI Ultra | $100/mo | Higher Gemini app, Antigravity, media, and agent limits; Gemini Spark beta starts in the U.S. | Check current plan page | Power users who need premium agent/media features |
| AI Ultra top tier | $200/mo | Google says the previous top-tier Ultra capabilities now cost $200/mo instead of $250/mo | Check current plan page | Heavy creators, developers, and agentic workflows |
API pricing via Gemini API pricing docs:
| Model | Input ($/MTok) | Output ($/MTok) | Context | Max Output |
|---|---|---|---|---|
| Gemini 3.1 Pro (older verified row) | $2-$4 | $12-$18 | 1M | 64K |
| Gemini 3.5 Flash | Check current docs | Check current docs | Check current docs | Check current docs |
| Veo 3.1 Lite | Per-video pricing | · | · | · |
Subscription prices and I/O changes verified 2026-05-20 via Google AI plans, Gemini subscription details, and Google’s May 19 subscription update. Gemini 3.5 API prices should be checked on the current Gemini API pricing page before production quoting.
Verification note: the older Gemini 3.1 Pro API rates remain useful historical context, but new Gemini 3.5 Flash / 3.5 Pro production budgets should be built from current Google API docs, not copied from this older row.
Against the alternatives
| Gemini 3.5 Flash / Google AI Pro | Claude Opus 4.7 / Claude Pro | ChatGPT Plus / Pro | |
|---|---|---|---|
| Best surface | Search, Workspace, Android, multimodal bundle | Long-form reasoning and terminal coding | General assistant plus Codex and GPT Image 2 |
| Agentic coding | Antigravity 2.0 and Managed Agents | Claude Code is stronger in terminal | Codex is stronger in OpenAI ecosystem |
| Image generation | Imagen / Nano Banana surfaces | None native | GPT Image 2 |
| Video generation | Veo / Flow / Gemini Omni surfaces | None native | None after Sora shutdown |
| Research/search | AI Mode in Search and Deep Research | Web search with citations | Broad browsing and research tools |
| Best viewed as | Google ecosystem AI layer | Reasoning and coding specialist | Broad generalist assistant |
Failure modes
- Plan complexity is still high. Free, AI Plus, AI Pro, AI Ultra $100, and AI Ultra $200 can be confusing. Verify exact feature limits in your own account before buying.
- Rollouts vary. Gemini Spark, Search agents, Omni, Antigravity features, and some media limits vary by region, language, plan, and release stage.
- Gemini 3.5 API pricing and limits need fresh checks. Do not reuse older Gemini 3.1 context or per-token pricing when quoting a new Gemini 3.5 production workload.
- Writing quality defaults to corporate-safe. Prose output is functional but often less distinctive than Claude. If voice matters, do not make Gemini the primary drafter without editing.
- Privacy posture differs by account type. Consumer Google accounts and Workspace/enterprise accounts do not carry the same data-handling guarantees. Sensitive document work belongs in managed Workspace or Google Cloud contexts.
- Model-name churn continues. Google has renamed plans and assistant surfaces repeatedly. Read the current subscription page before buying or publishing pricing.
Methodology
This page was produced by the aipedia.wiki editorial pipeline, an automated system that ingests vendor documentation, verifies pricing and model details against primary sources, and generates the editorial analysis you are reading. No individual human wrote this review. Scoring follows the four-dimension rubric at /about/scoring/ (Utility x Value x Moat x Longevity, unweighted average). Last verified 2026-05-22 against Google’s Gemini 3.5 announcement, the May 21 Gemini ADK and Home infrastructure update, the May 20 Marketing Live update, the May 19 I/O subscription update, the May 19 Antigravity and Managed Agents update, Google AI plans, Gemini subscriptions, Gemini API pricing docs, and the May 12 Android Show: I/O Edition coverage.
FAQ
What’s the difference between Gemini 3.5 Flash and Gemini 3.5 Pro? Gemini 3.5 Flash is the model Google launched broadly at I/O 2026 for the Gemini app, AI Mode in Search, Antigravity, and Gemini API surfaces. Google says Gemini 3.5 Pro is in testing and expected after I/O; do not assume Pro availability until it appears in your account or API docs.
Is Gemini free? Yes. The free tier gives limited Gemini access. Google AI Pro at $19.99/mo remains the mainstream paid tier. AI Ultra now starts at $100/mo, with a $200/mo top tier for heavier media, agent, Antigravity, and Gemini app limits.
How does Gemini compare to Claude and ChatGPT? Pick Gemini for Google ecosystem work: Search, Gmail, Docs, Drive, Android, YouTube, Google Cloud, and Antigravity. Pick Claude for long-form reasoning and Claude Code. Pick ChatGPT when you want the broadest general assistant, Codex, GPT Image 2, and OpenAI’s plugin-style ecosystem.
What is Deep Research and how does it compare to Perplexity? Deep Research is Gemini’s autonomous research agent. Perplexity is usually faster for iterative source-first research. Gemini is more valuable when the research output needs to stay inside Google’s broader assistant and Workspace ecosystem.
What changed at I/O 2026? Google launched Gemini 3.5 Flash broadly, made it the default model for AI Mode in Search, announced more agentic Gemini app features, expanded Antigravity and Managed Agents for developers, and reset Google AI Ultra pricing to $100 and $200 tiers.
Related
- Category: AI Chatbots · AI Writing · AI Search
- Google media tools: Gemini Omni · Google Veo 3.1 · Imagen 4
