← All Guides
Live Now
/
AI · Comparison
/
Published May 2026
Big Three
Ecosystems.
Claude, ChatGPT, and Gemini are not just different LLMs. They are three competing operating systems for how AI plugs into your work and your computer.
Each one has a desktop app, a terminal tool, a way to automate your browser, a research mode, a document editor, a file integration layer, and a developer platform. The terminology is completely different across all three. This guide is the Rosetta Stone.
Anthropic
Claude
Workspace folder, Cowork hub, document skills, MCP-native.
OpenAI
ChatGPT
Widest reach, Atlas browser, Agent Mode, GPT Store ecosystem.
Google
Gemini
Workspace-native, Spark agent, Chrome side panel, Nano Banana imagery.
///Contents
15 categories
///01
Engines under the hood
The Models.
Before the apps and products, there are the models. These are the actual AI systems doing the thinking. Everything else is interface.
Anthropic
Claude Opus 4.7
+ Claude Haiku 4.5
Sonnet 4.6 sits in the middle — the one you use most. Opus 4.7 launched April 16, 2026.
OpenAI
GPT-5.5 Pro / Thinking
+ GPT-5.5 Instant / 5.4 mini
Pro reserved for paid tiers. GPT-5.5 Instant became ChatGPT's default model on May 5, 2026.
Google
Gemini 3.1 Pro
+ Gemini 3.5 Flash
3.5 Flash launched at I/O 2026 — 4× faster than other frontier models. Gemini 3.5 Pro arrives next month.
Google also launched Gemini Omni at I/O 2026 — a model that takes image, audio, video, and text as input and outputs video. OpenAI's equivalent was Sora, but Sora was discontinued in April 2026. Anthropic doesn't have a video-generation model.
The Chat Interface.
The face of each AI. The thing your parents have heard of.
Anthropic
Claude.ai
claude.ai
Longer context windows and more careful reasoning.
OpenAI
ChatGPT
chatgpt.com
Widest user base — over 900 M users · $25 B ARR as of May 2026.
Google
Gemini
gemini.google.com
Most deeply integrated into the broader stack — Gmail, Docs, Drive.
All three work the same way in their simplest form: you type, it responds. The differences show up the moment you go one layer deeper.
///03
AI that runs while you work
The Desktop Agent.
This is where the product war is being fought in 2026. Each provider has a version of "AI that lives on your desktop, manages files, runs scheduled tasks, and ships work back to you."
Anthropic
Claude Cowork
Desktop app
Standalone desktop app with a dedicated workspace folder. Reads files directly, runs scheduled tasks, delivers outputs to a folder you own. Connectors, Skills, Plugins.
OpenAI
Agent Mode
Built into ChatGPT · Plus / Pro
Evolved from "Operator." Navigates websites, runs code, builds spreadsheets and slideshows, fills forms — from the same interface you chat in.
Google
Gemini Spark
Beta · Ultra subscribers
Announced at Google I/O May 20, 2026. 24/7 agentic assistant with Gmail, Docs, Workspace integrations. Runs on Google Cloud VMs — works when your laptop is closed.
/ The key difference
Cowork gives you a local file workspace and is the most mature product. Spark has the deepest first-party app integration (because Google owns Gmail, Docs, and Chrome). Agent Mode is the most seamlessly embedded into an existing conversation product.
///04
AI in your command line
Terminal · CLI.
All three providers have a coding agent you install via your terminal and run from the command line.
Anthropic
Claude Code
npm install -g
@anthropic-ai/claude-code
Reads and edits your codebase, runs commands, operates inside your git history.
OpenAI
Codex CLI
npm install -g
@openai/codex
75K+ GitHub stars, 14.5M monthly npm downloads. Goals, browser verification, plugins. Most widely adopted of the three.
Google
Antigravity CLI
npx @google/gemini-cli
Gemini CLI transitioning to Antigravity CLI. Free access ends June 18, 2026 — paid tiers retain access.
/ Not a technical person?
Installing a CLI means running commands directly on your device — proceed carefully. The commands above are a starting point, not a complete guide. Each provider has full installation documentation.
Computer Use.
All three have a mode where the AI takes control of your screen — clicking buttons, filling forms, navigating interfaces that don't have APIs.
Anthropic
Dispatch
Sees your screen, moves your cursor, operates any application. Designed for repetitive tasks in apps without an API. Slow by design — you pay attention to it.
OpenAI
Codex Computer Use
Operates macOS apps autonomously with its own cursor. Multiple Codex agents can run in parallel without disrupting foreground work. Currently one of the most capable implementations.
Google
Gemini Spark
Browser control through Chrome, with the ability to send texts and emails and operate your browser directly. Coming in the weeks after Spark's launch.
///06
Persistent workspaces
Projects.
A "project" is a saved context — a place where you store files, instructions, and conversation history related to a specific ongoing effort.
Anthropic
Claude Projects
Persistent memory, reference files, and custom instructions per project. Context retained across sessions.
OpenAI
ChatGPT Projects
Group related chats, upload reference files, add custom instructions. The AI stays on-topic to the project context.
Google
Notebooks · NotebookLM
Two-pronged. Notebooks in Gemini is project management synced with NotebookLM — source-grounded research with Gemini Deep Research integrated.
Document Creation.
Beyond generating text in a chat window — AI that produces actual editable documents, spreadsheets, presentations, and interactive dashboards.
Anthropic
Skills + Live Artifacts
Word · Excel · PPT · PDF
Skills handle formatted document creation — real .docx, .xlsx, .pptx files. Live Artifacts (April 21, 2026) are interactive HTML pages that pull fresh data every time you open them.
OpenAI
ChatGPT Canvas
Collaborative document editor inside ChatGPT. As of 2026, can also build interactive apps and call APIs — more powerful than a simple editor.
Google
Docs Live + Workspace AI
Docs Live (I/O 2026) — conversational interface for creating and editing documents directly in Docs. Rolling out to AI Pro and Ultra on iOS and Android.
Image Generation.
Three providers, three completely different bets on what generative imagery is for.
Anthropic
Claude Design
Dedicated product for interfaces, slides, diagrams, and interactive visualizations through Artifacts. Research preview April 17, 2026, powered by Opus 4.7. Image analysis still considered best-in-class.
OpenAI
GPT-Image-2
Formerly DALL-E
DALL-E 3 retired March 2026. GPT-Image-2 (April 21, 2026) now default inside ChatGPT — reasoning before generation, 4K resolution, multilingual text rendering.
Google
Nano Banana
Nano · Pro · 2
Three tiers: original (Aug 2025), Pro (built on Gemini 3 Pro), and Nano Banana 2 (Feb 2026) — default in Search, Lens, AI Mode across 141 countries. 4K, multi-image blending, SynthID watermark.
/ The key difference
Google is the clear leader here with a mature, multi-tier product embedded across its ecosystem. OpenAI has a capable replacement in GPT-Image-2. Claude focuses on design and structured visuals over photorealistic rendering.
Deep Research.
Multi-step research mode — ask a big question, the AI browses the web autonomously, synthesizes sources, returns a full report.
Anthropic
Research mode
Built into Claude. Synthesizes sources and structures a report alongside your conversation.
OpenAI
Deep Research
One of the defining product launches of 2025. Finds, analyzes, and synthesizes hundreds of online sources to produce a research-analyst-level report. Available on Plus, higher limits on Pro/Enterprise.
Google
Gemini Deep Research
Embedded inside NotebookLM. Combines Google's search grounding with source-cited research — run a query, immediately work with the sources in your Notebook.
///10
AI inside your browser
The Browser.
Three completely different bets — an extension, a side panel, or a whole new browser.
Anthropic
Claude in Chrome
Chrome extension
Brings Claude into any webpage: summarizing content, drafting replies, running Claude while you browse. Installed from the Chrome Web Store.
OpenAI
ChatGPT Atlas
Standalone browser · macOS
A Chromium-based browser (October 2025) with ChatGPT in the sidebar, an agent mode that can operate websites, and browser memory across sessions.
Google
Gemini · native in Chrome
Persistent side panel
Built directly into Chrome. Auto Browse (Jan 2026) is an agentic mode that can scroll, click, fill forms, and navigate websites for you. AI Pro and Ultra.
/ Note on browser tools
Installing extensions or dedicated AI browsers carries inherent security risks. Only install from official sources, review permissions, and proceed at your own discretion.
///11
AI in your spreadsheet
Workspace Ext..
Each provider has a way to bring AI into the spreadsheet you already use.
Anthropic
Claude in Excel
Add-in · Microsoft AppSource
Writes formulas, analyzes data, builds spreadsheets through conversation. Requires a paid Claude plan. Web, Windows, Mac (16.46+).
OpenAI
ChatGPT for Excel
Via Microsoft Copilot
Exists through Microsoft's broader Copilot and OpenAI partnership within Microsoft 365.
Google
Gemini in Google Sheets
Native
Native. Gemini is already inside every Google Workspace product.
///12
How AI plugs into your tools
Connectors.
Live integrations to the third-party tools you already use.
Anthropic
Claude Connectors
GitHub, Slack, Gmail, Drive, Notion, Asana, Linear, Jira, Figma, Stripe, Canva, ClickUp, and dozens more. All connections use MCP.
OpenAI
App Integrations
Pilot partners include Figma, Canva, Booking.com, Coursera, Expedia, Spotify, Zillow. Available to all logged-in users on Free, Go, Plus, Pro (outside EEA, CH, UK for now).
Google
Google MCP + Workspace
Home-team advantage — Gmail, Drive, Docs, Sheets, Calendar are all native. External tools via Google's MCP. Spark launched with Canva, OpenTable, Instacart.
/ Terminology
MCP is the protocol — the plumbing that makes connections possible. A Connector is a single tool integration built on MCP. A Plugin bundles multiple Connectors and Skills into one installable package.
///13
Reusable AI configurations
Skills · GPTs · Gems.
Saved AI setups tuned for a specific job — a customer support persona, a coding reviewer, a brand voice.
Anthropic
Skills + Plugins
Instruction bundles for specific tasks. Install community Skills or build your own. Plugins bundle MCPs and Skills into one installable package from a plugin marketplace.
OpenAI
Custom GPTs
User- or community-created AI configurations anyone can publish. The GPT Store has millions. Think apps built on top of ChatGPT.
Google
Gems
Gemini's equivalent — saved AI personas with custom instructions and tools attached.
Developer Platform.
All three have full APIs for building products on top of their models. The interesting layer in 2026 is the agent SDK — tools for building multi-step autonomous AI agents.
Anthropic
Claude Agent SDK
+ Anthropic API
Powers Cowork under the hood. Framework for building agents that use tools, manage memory, and execute multi-step workflows. Available to build Cowork-style functionality into your own products.
OpenAI
OpenAI Agent SDK
+ OpenAI API
Foundation of the Codex agent platform. Supports tool use, parallel agents, and code execution.
Google
Antigravity SDK
+ AI Studio · Gemini API
Announced at I/O 2026. Programmatic access to the same agent harness powering Google's own products. Co-optimized for Gemini models, self-hostable.
///15
What you're actually paying for
Subscription Tiers.
All three power-user tiers converge on the same $100–$200 range. Each splits into two sub-tiers, with the higher offering roughly 4× the usage.
Anthropic
OpenAI
Google
Free
Claude.ai (limited)
ChatGPT Free
Gemini Free
Low-cost paid
—
ChatGPT Go · $8/mo
Google AI Plus · $7.99/mo
Basic paid
Claude Pro · $20/mo
ChatGPT Plus · $20/mo
Google AI Pro · $19.99/mo
Power user
Claude Max · $100–200/mo
ChatGPT Pro · $100–200/mo
Google AI Ultra · $100–200/mo
Teams
Claude Team
ChatGPT Team
Google AI for Workspace
Enterprise
Claude Enterprise
ChatGPT Enterprise
Gemini Enterprise
As of May 2026.
The Full Map.
Category
Claude
OpenAI
Google
Main chat
Claude.ai
ChatGPT
Gemini
Desktop agent
Claude Cowork
ChatGPT Agent Mode
Gemini Spark (beta)
Terminal / CLI
Claude Code
Codex CLI
Antigravity CLI
Computer use
Dispatch
Codex Computer Use
Gemini Spark (coming)
Persistent projects
Claude Projects
ChatGPT Projects
Notebooks / NotebookLM
Document creation
Skills (Word / Excel / PPT / PDF)
Canvas
Docs Live / Workspace
Image generation
— (Claude Design)
GPT-Image-2
Nano Banana / Pro / 2
Live dashboards
Live Artifacts
Canvas (app mode)
—
Deep research
Research mode
Deep Research
Gemini Deep Research
Browser extension
Claude in Chrome
ChatGPT Atlas
Native in Chrome
Spreadsheet ext.
Claude in Excel
ChatGPT for Excel
Gemini in Sheets
Third-party integrations
Connectors (via MCP)
App Integrations
Google MCP + Workspace
Saved AI configs
Skills + Plugins
Custom GPTs
Gems
Developer API
Anthropic API
OpenAI API
Gemini API / AI Studio
Agent builder framework
Claude Agent SDK
OpenAI Agent SDK
Antigravity SDK
///Compass
Centres of gravity
How to Think About It.
Anthropic / Claude
Serious work, embedded.
Building for people who want AI embedded in serious work: document creation, project management, scheduled automation, creative professional workflows. Cowork is the hub. Everything connects back to a workspace folder you own.
OpenAI / ChatGPT
Widest reach, most agentic.
Building for the widest possible audience and the most deeply agentic product. Codex Computer Use running multiple agents in parallel is genuinely impressive. The GPT Store and massive third-party ecosystem mean there's a ChatGPT for almost every niche task.
Google / Gemini
Already in your world.
Building for people already inside Google's world. If you live in Gmail, Docs, Calendar, Spark is the most seamless path — it doesn't have to ask for permission. Antigravity SDK and Spark represent a major bet on agents.
The browser wars ended with Chrome winning. This one is still wide open.
///Follow along
Want the next one?