Gemini in 2026: Google's Full AI Stack Explained (Chrome, Workspace, CLI, AI Mode, Antigravity)

If you only follow Anthropic and OpenAI announcements, you can miss the size and shape of Google's AI stack. Gemini now spans Search, Chrome, Workspace, the terminal, a desktop agent app, the IDE and the developer API, all sharing one model family but each shaped for a different surface. The biggest change this year is in the terminal: Google retired Gemini CLI for individual users on June 18, 2026 and moved them to Antigravity CLI. This guide is the map: what each piece is, when it wins, and how to choose between Gemini and the alternatives.

If you are evaluating Google's coding tools specifically against Claude Code, Codex CLI, and Cursor, that comparison is the right starting point. This article is the broader stack view: where Gemini lives across your whole computing surface, not just your editor.

Key Takeaways

• Two model lines do most of the work: Gemini 3.1 Pro (still a preview model) and Gemini 3.8 Flash, which Google calls its most intelligent Flash model.
• The terminal agent is now Antigravity CLI. Gemini CLI stopped serving free, Google AI Pro and AI Ultra users on June 18, 2026; enterprise Code Assist licenses and paid API keys keep it.
• Antigravity grew into a platform: a desktop app (Antigravity 2.0), a CLI, an IDE and an SDK, generally available with a $0 individual plan.
• Where Google wins: video and audio input, search-grounded answers, Workspace integration, and Chrome as a surface.
• Where Google is not the default: our model picks put Claude Opus 5.5 and GPT-6 Astra first and second; Gemini 3.1 Pro is our pick for video and audio input.

Google Gemini Plugin and Tool Definition Patterns, 2024-2026

Google Gemini plugin and tool definition patterns split by surface. In Chrome, a "tool" is usually a saved Skill: a reusable prompt with the current page or selected tabs as context. In Antigravity, the tool pattern is developer-agent infrastructure: Agent Skills, hooks, subagents, MCP servers and plugins, the features Google says Antigravity CLI carried over from Gemini CLI. In the Gemini API, tools are explicit callable functions with structured arguments that your application decides how to execute.

The important shift from 2024 to 2026 is that Gemini tools are no longer one generic plugin system. They are surface-native. A Chrome Skill is intentionally simple because it has to work for non-developers. Antigravity's skills and hooks are explicit because developers need repeatable tool contracts, and its artifacts (plans, screenshots, browser recordings, diffs) are what you review. The Gemini API is the lowest-level pattern: you define the tool, the model requests it, and your app controls execution.

If you are trying to decide which tool pattern to copy, start with the user's surface. Browser readers need Chrome Skills. Developers need Antigravity or Gemini API tool definitions. Teams orchestrating multiple coding tasks need Antigravity 2.0. The model matters, but the tool contract and review surface decide whether the workflow is actually useful.

Map of the Stack

The cleanest way to think about Google's AI footprint is the surface it lives on:

You can use any of these without the others. Most teams use two or three: Workspace + AI Mode if you live in Google's productivity stack, Antigravity if you live in code, the API if you build products.

Gemini 3.1 Pro and Gemini 3.8 Flash: The Models Powering Most of It

Google's model page lists Gemini 3.1 Pro, released February 19, 2026, as its Pro model, still in preview, and Gemini 3.8 Flash as a stable model it calls "our most intelligent Flash model." The headline numbers from Google's model and pricing pages:

• Context: both take 1,048,576 input tokens and return up to 65,536 output tokens.
• Multimodal input: text, image, video, audio and PDF.
• Gemini 3.1 Pro pricing (standard tier): $2 per million input tokens up to 200K context, $4 above 200K; $12 per million output tokens up to 200K, $18 above. Batch runs at half those rates.
• Gemini 3.8 Flash pricing: $0.75 per million input tokens and $3.75 per million output tokens through December 31, 2026, rising to $1.50 and $7.50 on January 1, 2027.
• Reasoning control: the thinking_level parameter lets you trade quality for cost per call, the same pattern as Claude's effort levels and OpenAI's reasoning effort.
• Endpoints: gemini-3.1-pro-preview, gemini-3.1-pro-preview-customtools for workflows that lean on bash and custom tools, and gemini-3.8-flash.

A million-token window is no longer unique to Gemini: our model picks page notes that Claude Opus 5.5 and GPT-6.1 Sol also take about a million tokens. What still sets Gemini 3.1 Pro apart is input type. It reads recordings and video natively, which is why it holds the "best for video and audio input" slot on that page.

Gemini in Chrome: The Side-Panel + Skills

Gemini in Chrome is a side-panel chat that knows what page you are on. Open the panel with the sparkle icon (or Ctrl+Shift+/ on Windows / Cmd+Shift+/ on Mac), and Gemini reads the current page as context. Ask "summarize this", "extract all the prices into a table", "draft a Slack message about this article": it works on the page in front of you with no copy-paste.

The April 14, 2026 launch added Chrome Skills: saved Gemini prompts you can run with one click. A set of pre-built Skills ships with it, browsable at chrome://skills/browse; Google publishes neither a count nor a category list, so treat any inventory you read elsewhere as a snapshot of someone's browser. You can save your own prompts as Skills, edit any pre-built one, and, in Google's words, a Skill "will run on the page you're viewing, along with any other tabs you select."

Use Chrome Skills when:

• You catch yourself typing the same prompt against different web pages more than twice a week. Save it, name it, and trigger it with one click forever after.
• You want a casual reader (someone less comfortable with prompt engineering) to get pro-level workflows without writing prompts. The pre-built Library covers most common patterns.
• You read across multiple tabs and want the same operation applied to all of them (price comparison across product pages, summary across multiple article tabs, extract dates from a set of meeting links).

Limits to know about: Chrome Skills is desktop-only, Chrome's help page requires a Chromebook Plus, Mac, or Windows computer, the browser language must be English (United States), and you need a personal Google account. Managed and Enterprise accounts are excluded unless an administrator enables the feature. It is free. Skills are not read-only: Google says a Skills prompt "will ask for confirmation before taking certain actions, such as adding an event to your calendar or sending an email."

For a deeper how-to, see our Chrome Skills complete guide. For what Google does and does not document about the pre-built library, see Chrome Gemini Skills Library: What Google Documents.

Gemini in Workspace: The Productivity Layer

Gemini in Workspace is the integration that lives inside Docs, Sheets, Gmail, Slides, Drive, and Meet. The product is shaped around the work you already do in those apps:

• Gemini in Docs: "Help me write" suggestions inline; full-document summaries; rewrite passages in different tones; turn meeting notes into action items.
• Gemini in Sheets: Generate formulas from natural language; analyze ranges; auto-classify rows; draft pivot tables.
• Gemini in Gmail: Inbox summary; "Help me write" for drafts; thread summaries; suggested replies that pull context.
• Gemini in Slides: Generate slides from a doc or a prompt; image generation inline; speaker notes from an outline.
• Gemini in Meet: Live note-taking with action items; "Take notes for me" recordings with summaries; instant translation.

Workspace's edge is that none of this requires a context switch. The agent is in the app you are already in, with read access to the document you are already on. For teams that live in Google Workspace, this is the highest-leverage Gemini product because the friction is zero.

The trade-off is that Workspace integrations are less powerful than dedicated tools for the same task. ChatGPT or Claude in a separate tab can do more sophisticated work with the same data because you control the prompt entirely. Workspace integrations are optimized for the 80% case, not the 20% power-user case.

Pricing: Google's Workspace pricing page lists Business Starter at $7 per user per month with the "Gemini AI assistant in Gmail" and chat with Gemini, and Business Standard at $14 with the "Gemini AI assistant in Gmail, Docs, Meet, and more." Introductory discounts apply for new customers.

The weekly AI Catchup

Get the week's AI coding news, distilled.

One email each week: the tools, releases, and workflows that matter for Claude Code, Codex, Cursor, and the agent stack.

Subscribe

From Gemini CLI to Antigravity CLI: The Terminal Agent

Google's terminal agent for individual developers is now Antigravity CLI. On May 19, 2026, Google announced that it was unifying its developer tools into Antigravity, and set the cutoff: "On June 18, 2026, Gemini CLI and Gemini Code Assist IDE extensions will stop serving requests for Google AI Pro and Ultra, as well as those using it free of charge using Gemini Code Assist for individuals."

Google's own pages disagree on one point. The Gemini CLI quota page opens with a banner saying "Unpaid tier and Google One users: Gemini CLI was replaced by Antigravity CLI on June 18th, 2026," but its table still lists 1,000 requests a day for Gemini Code Assist (Individual), 1,500 for AI Pro, 2,000 for AI Ultra and 250 for an unpaid API key. Treat the banner and the blog post as current; the table has not been updated to match.

What Antigravity CLI keeps and changes, per Google:

• Carried over: Agent Skills, Hooks, Subagents, and Extensions (now Antigravity plugins). Google says there "won't be 1:1 feature parity right out of the gate."
• Rebuilt in Go, which Google says makes it "snappier and more responsive."
• Background agents: it orchestrates multiple agents for large tasks without locking up your terminal.
• One harness with the desktop app: Antigravity CLI and Antigravity 2.0 share the same agent harness and settings, and you can pull a CLI conversation into the desktop app.
• Migration: Google's Antigravity docs include a Gemini CLI migration guide for moving skills, MCP servers and other customizations.

Gemini CLI itself is still an open-source project on GitHub under Apache 2.0, and its docs still cover headless mode, MCP servers, plan mode, sandboxing and GEMINI.md project context. That matters if your organization holds a Code Assist license or runs it on a paid API key; for everyone else, new terminal work belongs in Antigravity CLI.

Use Antigravity CLI when:

• You want a terminal agent on Google's free individual plan.
• Your work is tied to Google Cloud; Google says you can use Antigravity CLI with your Google Cloud projects.
• You need long inputs such as whole codebases, recordings or video alongside code, where Gemini's multimodal input helps.
• You also use Antigravity 2.0 and want the same settings and permissions in both places.

For the architectural trade-offs against Claude Code and Codex CLI, see our Codex CLI vs Claude Code vs Cursor architecture comparison.

Google Antigravity: Desktop App, CLI, IDE and SDK

Antigravity is Google's agentic development platform, first released in public preview on November 18, 2025 as an "agent-first IDE." On May 19, 2026, Google expanded it into four products, and its pricing page now labels the individual plan "Generally Available":

• Antigravity 2.0. A standalone desktop app with no IDE, which Google calls "your command center to manage multiple local agents in parallel." It keeps the principles of the original Agent Manager and adds dynamic subagents, asynchronous tasks, JSON hooks, Scheduled Tasks that run agents on a cron, and projects that span several folders, per Google's launch post.
• Antigravity CLI. The terminal surface on the same harness (see above).
• Antigravity IDE. The full agentic IDE "complete with the agent manager, artifacts, and a deep understanding of your codebase." Artifacts are implementation plans, screenshots, browser recordings and walkthroughs you review instead of reading raw logs.
• Antigravity SDK. Python scripts for building custom agents and running evaluations on the Antigravity harness.

It runs on macOS, Windows, Linux, and Googlebook.

Models. Antigravity's models page lists Gemini 3.8 Flash, 3.7 Flash, 3.6 Flash and 3.1 Pro on every plan. Claude Sonnet 5.5 and Claude Opus 5.5 are available on Google AI Pro (non-trial subscriptions only) and AI Ultra, but not on the free plan or Enterprise. Claude Sonnet 4.6, Claude Opus 4.6 and GPT-OSS-120b are still listed but will be removed on November 2, 2026.

Plans. The individual plan is $0 a month with "Basic weekly rate limits." Google AI Pro and AI Ultra raise the limits and add a flexible AI credit pool. In May, Google added a $100 a month AI Ultra plan alongside the $20 Pro plan and cut its top Ultra plan from $250 to $200, with rate limits of 5x and 20x the Pro plan. Google Cloud customers get Antigravity through Gemini Enterprise.

Use Antigravity when:

• You are running multiple parallel coding tasks and want a dashboard rather than a chat thread.
• You want agents to run on a schedule from the desktop app.
• You want to mix Gemini models with Claude Sonnet 5.5 or Opus 5.5 on a paid Google AI plan.
• You are open to a new IDE or desktop app rather than committing to Cursor.

AI Mode and AI Overviews in Google Search

This is the surface most readers encounter without realizing it is part of Google's AI stack. AI Overviews is the summary box at the top of search results, generated from indexed pages and cited as it goes. AI Mode is the separate conversational tab where you can have a multi-turn search-grounded conversation rather than typing one query at a time.

For most readers these are not "products you use"; they are features of Google Search you encounter. If you publish content yourself, the practical lesson is that both surfaces cite pages they can extract from: answer-first paragraphs, a clear H2/H3 hierarchy, and schema markup are the same habits that help in classic organic search.

The Gemini API: For Developers Building Their Own Surfaces

The Gemini API is what you call when you build Gemini into your own product. Two paths:

• Google AI Studio + Gemini Developer API. Lower friction, individual developer focus, free tier for prototyping (Gemini 3.8 Flash has a free tier; Gemini 3.1 Pro does not).
• Vertex AI. Google Cloud-integrated, enterprise focus, with the full GCP IAM/billing/observability story.

The differentiator is integration: if your product already runs on Google Cloud, Vertex AI is significantly less friction than wiring up Anthropic or OpenAI side-by-side.

Use cases that lean toward Gemini specifically:

• Video and audio input (meeting recordings, screen captures, podcasts), where Gemini takes the file directly.
• Multimodal pipelines (PDF + image + audio in one request).
• Search-grounded responses. Grounding with Google Search includes 5,000 free search requests a month shared across Gemini 3.x models, then $14 per 1,000.
• Cost-sensitive workloads at scale, where Gemini 3.8 Flash's $0.75 input price applies through the end of 2026.

How the Pieces Fit Together: Three Real Workflows

The stack-level picture is clearest with concrete combinations:

Workflow 1: The Workspace-First Operator

Spend most of your day in Docs, Sheets, and Gmail. The Gemini stack you actually use:

• Gemini in Workspace for inline drafting, summarizing, and transforming inside the apps.
• Gemini in Chrome with Skills for one-click research workflows on the web.
• Gemini Apps (gemini.google.com) when you want a longer conversation outside any specific document.
• AI Mode + AI Overviews as your default search surface.

You probably do not need Antigravity or the API. Workspace Business Standard at $14 per user per month covers Gemini across Gmail, Docs and Meet.

Workflow 2: The Solo Developer

Building software, mostly alone, want to keep costs low while still using AI agents heavily:

• Antigravity CLI as your terminal agent on the free individual plan, upgrading to Google AI Pro when you hit the weekly limits.
• Antigravity 2.0 for bigger refactors and scheduled agent runs.
• Gemini in Chrome for documentation lookup while coding.
• Gemini API for product features, starting on Gemini 3.8 Flash's free tier.

Workflow 3: The Multi-Vendor Power User

You want the best of every vendor and you do not mind paying for it:

• Claude Code for daily coding work, with Claude Opus 5.5, our top-ranked model.
• Cursor for visual editing and the marketplace.
• Antigravity for parallel agent runs, with Gemini and Claude models side by side on a Google AI plan.
• Gemini API for video and audio input and high-volume Flash calls.

This mix costs more, but it gives you access to every important capability in the ecosystem.

How to Decide Between Gemini and the Alternatives

A simple decision tree:

• Need a free option? → Antigravity's $0 individual plan or Gemini Apps.
• Live in Google Workspace? → Gemini in Workspace beats any external alternative on integration friction.
• Need video or audio input? → Gemini 3.1 Pro. GPT-6.1 Sol accepts text and images only.
• Need a million-token window? → Gemini 3.1 Pro, Claude Opus 5.5 and GPT-6.1 Sol all take about a million tokens; pick on price and task.
• Need agentic coding workflows? → Claude Code or Codex CLI for terminal, Cursor or Antigravity for IDE.
• Need search-grounded answers in your product? → Gemini API with grounding is the cleanest path.
• Building a product on Google Cloud? → Vertex AI Gemini is the lowest-friction option.

What This Means for the Next Six Months

Three observations from where Google's stack stands now:

• Google consolidated its developer tools. Gemini CLI and the Gemini Code Assist IDE extensions gave way to one Antigravity harness shared by a desktop app, a CLI, IDE extensions and an SDK. Expect new agent features to land there first.
• Antigravity is now a multi-model platform. It ships Gemini Flash releases within weeks (3.6, 3.7 and 3.8 Flash between July and September 2026) and offers Claude Opus 5.5 on paid plans.
• The model lead is spread across labs. Our model picks rank Claude Opus 5.5 first overall, GPT-6 Astra second as the auditor, and GPT-6.1 Sol third for high-volume coding agents, with Gemini 3.1 Pro the pick for video and audio input. Stay multi-vendor and let the surface you commit to outlast any single model lead.

The biggest practical takeaway: Google's AI stack is bigger than most developers outside Google shops realize. If you used Gemini CLI and stopped when it went away, Antigravity CLI and Antigravity 2.0 are where that work continues, and the free individual plan makes an afternoon of testing cheap.

Corrections and Updates

• October 9, 2026: This page said Gemini CLI's free tier gave 60 requests per minute and 1,000 a day and called Antigravity a public preview running Gemini 3 Pro and Claude Sonnet 4.5; Gemini CLI stopped serving free, AI Pro and AI Ultra users on June 18, 2026, and Antigravity is now generally available with Gemini 3.8 Flash and Claude Opus 5.5.
• October 9, 2026: This page said Workspace Business Starter does not include Gemini, and cited AI Overviews figures (76% overlap with top-10 results, 47% of citations from outside the top 50) without a source; Google lists the Gemini assistant in Gmail on Starter, and the unsourced figures are removed.

Sources

• Google Developers Blog, "An important update: Transitioning Gemini CLI to Antigravity CLI", May 19, 2026. The June 18, 2026 cutoff, who it covers, enterprise access, paid API keys, and the features Antigravity CLI carries over. Checked October 9, 2026.
• Gemini CLI docs, "Quotas and pricing", last updated June 18, 2026. The replacement banner and the quota table that still lists daily limits. Checked October 9, 2026.
• google-gemini/gemini-cli on GitHub. Apache 2.0 license; repository still active. Checked October 9, 2026.
• Google Antigravity and Antigravity pricing. The four products, platforms, the $0 individual plan, "Generally Available," and Google Cloud access. Checked October 9, 2026.
• Antigravity docs, "Models". Models by plan and the November 2, 2026 removals. Checked October 9, 2026.
• Antigravity blog, "Introducing Google Antigravity 2.0", "Google Antigravity CLI" and "Changes to Antigravity plans", May 19, 2026. Antigravity 2.0 features, the shared harness, and the $20, $100 and $200 plans. Checked October 9, 2026.
• Antigravity blog, "Introducing Google Antigravity", November 18, 2025. The original public preview.
• Gemini API docs, "Models" and Gemini 3.8 Flash model page. Model lineup, preview status, token limits and input types. Checked October 9, 2026.
• Gemini API pricing. Gemini 3.1 Pro and Gemini 3.8 Flash prices, free tiers, and grounding prices. Checked October 9, 2026.
• Google Workspace pricing. Business Starter and Business Standard prices and Gemini features. Checked October 9, 2026.