Build apps, sites, and even Chrome extensions in natural language; Artifacts render and run live in the chat.

Describe a site or app and it builds, hosts, and publishes it at a shareable URL, with versioning and access controls. Public beta on paid plans; not on the free or entry tiers.

Turns prompts into working apps, games, infographics, quizzes, and web pages you can refine and export.

Labs can spin up mini-apps and dashboards from research, but it isn't built for real coding work.

Writes and debugs code well in the conversation, but there's no live build-and-preview surface — you copy the code out to run it.

An agent mode that builds multi-page websites and editable slide decks from a prompt; capable but still beta and best for first drafts.

Describe a website, app, game, or dashboard and Grok builds a working preview in the chat, then publishes it to a grok.me link or your own domain. Still labeled early beta; output quality reported as uneven.

Builds and hosts working sites and apps from a prompt, with one-click publishing to web, iOS and Android. A separate Game Dev environment builds playable, shareable games, including multiplayer ones.

Creates runnable apps as artifacts. It can also pass work to an app-building tool through a connector (Lovable is on the connector list).

Builds real .xlsx, .pptx, .docx, and PDF files from a chat, with formulas and formatting — free for all users, 30MB per file.

Generates spreadsheets and files via its data-analysis engine, plus a Microsoft add-in covering Word, Excel, and PowerPoint that is available on all plans including free; strongest for code-related exports.

Creates Docs, Sheets, Slides, and PDFs directly in chat, free for all users since April 2026.

Labs turns research into slides, spreadsheets, and dashboards, but it's a Pro/Max feature, not a general file maker.

Generates Word/Excel/PowerPoint natively — it lives in Office — and Notebooks can build a full deck; needs a paid Office/Copilot plan.

Generates slide decks, spreadsheets, and documents exportable to Office formats through its Work tools on paid tiers.

Built-in Skills generate formatted .docx, .pptx, .xlsx, and PDF files, and can restructure files you upload. Included with every account.

Produces Word, Excel, PDF and native editable PowerPoint files (PowerPoint in beta). Formatting can be inconsistent on complex outputs.

Creates PDFs, DOCX, TXT, Markdown, XLSX, CSV, PNG, JPEG, GIF, WebP, code files, browser-viewable web tools, audio, and podcasts, saved to the Library tab.

Set instructions and upload files per project; Skills and Plugins add packaged capabilities. Free users get up to 5 projects.

A large ecosystem via the GPT Store. Build with instructions, knowledge files, and actions. Browsing is free; building needs a paid plan.

Custom personas with instructions and uploaded files, now free to create and shareable via Drive links. They plug into Workspace apps like Docs and Gmail.

More a research workspace with custom instructions and files than a shareable bot. File uploads require a Pro plan.

The consumer app has almost no bot-builder. Real customization lives in Copilot Studio, which is enterprise, low-code, and separately priced.

A marketplace of pre-built specialized agents (academic writing, coding, presentations) plus 'Document-to-Skills' for turning documents into reusable skills; less a from-scratch bot builder.

Save reusable instructions, formats, and reference files as a Skill that Grok applies in any chat; build one by describing it or with the built-in Skill Creator. There is no standalone, shareable assistant with its own chat page; named persistent assistants live only in the separate Grok Bot app. Available on Free; paid plans add usage headroom.

No standalone assistant builder. Projects apply reusable instructions and knowledge files to every new task, while Skills add reusable capabilities. Manus's separate Cue app creates personal agents but sits outside this product.

Persistent workspaces with files, custom instructions, and project memory. Available even on the free tier (up to 5).

Groups related chats, files, and instructions under one project.

Notebooks give each topic its own space for chats, files, and custom instructions, with notebook memory, and sync with Gemini Notebook (renamed from NotebookLM in July 2026). Free on personal accounts; extended to Workspace and Education accounts in September 2026.

Purpose-built for this, with team collaboration and unified search across your files and the web.

Copilot Notebooks are emerging but thin on the consumer side compared with the others.

Persistent workspaces that hold reference files, project instructions, and related chats, with context carried into every chat started inside them. Quotas scale by plan, starting at 2 projects and 500MB of storage. The Kimi Work desktop app has its own separate projects that don't share data with these.

Group chats, uploaded files, and custom instructions by topic so every chat in the project references them. Available on Free; paid plans add usage headroom. Some users have reported Projects missing when using grok.x.ai instead of grok.com.

Projects hold a master instruction plus a knowledge base of files that every new task inherits. On Team plans, members of a shared Project see only the tasks they created.

Generates documents, pages, charts, and interactive pieces you can iterate on live. Docs and Slides, added September 2026, add in-place editing, comments, and export to PowerPoint or PDF; both are in beta on paid plans.

Still there for writing and code, but folded back into the main chat window on the newest models.

A dedicated space to draft and refine documents or code, exportable to Google Docs.

Turns research into shareable published pages; more about publishing than collaborative editing.

Collaborative Copilot Pages for building and refining content with suggested edits.

Generates editable slide decks with an in-browser editor, though the editor is basic and you'll often export to PowerPoint for final polish.

The desktop workspace lets you edit what Manus made (docs, slides, sites, video timelines, game assets) by hand and hand it back. It's an editing environment more than live co-writing. On the free plan it's available but limited by the credit cap.

Searches the web when needed, but the web isn't the center of the product.

Web results with source links folded into a general assistant.

Answers grounded in Google Search, which helps on factual, up-to-date questions.

Its whole identity: citation-first answers, and now a full AI browser that's free across platforms.

Built into Edge and can look at what's on your screen or camera to help in context.

Searches the live web for current questions, but citations are lighter and there's no browser.

Web search and browsing are core and free; on paid tiers the WebBridge agent actively navigates sites for you.

Pulls live web results and live posts from X into answers. There is no Grok-branded AI browser; web and X search are limited on the free plan.

Works in a Manus-hosted browser that keeps you signed in, or through an extension in your own Chrome or Edge. Hosted browsing often hits CAPTCHAs that need you to take over.

Each user's agent has its own browser in an isolated cloud machine. A separate gatekeeper called Sentinel approves outbound actions and asks you before anything sensitive.

Multi-source research with citations, plus connectors to pull from your own tools.

Produces long, well-structured, cited reports at roughly junior-analyst quality.

Autonomous cited reports, now available to try for free, and can turn findings into interactive content in Canvas.

Fast, citation-dense, one-click source verification. Labs can output slides, sheets, and dashboards directly.

Consumer Deep Research was removed in the August 2026 Copilot app merger; its replacement, Researcher, produces multi-source cited reports but is limited to paid Microsoft 365 Premium.

Researches across sources and produces briefs and slides, though it can occasionally cite sources that don't exist — verify before relying on it.

A dedicated mode that runs multi-step searches across the web and live X posts and returns a cited report. It can appear on Free, but Free's search allowance is limited; once you hit the cap, it pauses or falls back to regular chat. Sources can mix social posts with published reporting, and independent tests have found citations that don't match the linked source, so check key claims.

Splits a large research job across many sub-agents that run in parallel and compiles the results into a document or spreadsheet. Sites that block automated browsing can leave gaps.

Researches on the web and in connected apps when a task needs it, and cites sources if asked. There is no dedicated report mode.

Native, high-quality image creation and editing inside the chat.

Strong image generation and editing, including mixed text-and-image responses.

Can produce images, but it's a side feature, not a strength.

Built-in, free image creation and editing in the Copilot app, with higher-quality options inside PowerPoint on premium plans.

No standalone image generation in chat, but the Slides tool generates per-slide imagery through an integrated third-party model.

Creates and edits images from text, with multi-image references. Removed from the free plan in March 2026; every output carries a visible Grok watermark that can't be turned off.

Generates images as a standalone task or as assets for slides, sites and videos. Each generation uses credits.

Generates images with Muse Image when a task calls for one.

Text-to-video with synced audio in the app, on paid plans. As of 2026 this is the only in-chat video generator here, after OpenAI discontinued Sora in April 2026.

Generates short video clips with native audio from text or images. Resolution is tier-gated: 480p on the entry plan, 720p on SuperGrok, 1080p on SuperGrok Plus and up; 720p requests fall back to 480p once you hit your cap.

Makes short videos (ads, tutorials, vlogs) and hands back an editable timeline, so you can swap clips or music without regenerating everything. Alchemy mode takes on creative direction. Available on the free plan, though restricted by credit caps.

Save a prompt to run on a recurring cadence.

Schedules recurring prompts like reminders and briefings — useful, but closer to a reminder tool than a full agent. Free accounts get up to three active tasks, running no more than once a day.

Runs recurring prompts, strongest when the task lives inside Google Workspace.

Recurring searches and research tasks that run on their own; heavier automation sits on the top tiers.

Recurring automation runs through agent flows and Cowork rather than a simple saved-prompt scheduler; enterprise-oriented.

Saved prompts that run on a recurring schedule, with a per-plan quota starting at 2 on the free tier. Schedules set in the Kimi Work desktop app run locally, so they don't fire while the machine is asleep or the app is closed.

Save a prompt to run once, daily, weekdays, weekly, monthly, or yearly. It can also run when a matching email arrives, but that trigger needs SuperGrok.

Runs on a schedule, and since Manus 2.0 can also start on events in connected services (new email, calendar event, Slack message, Notion update). Free accounts are limited to 2 scheduled tasks; advanced Automations are in beta.

Keeps working in the background and lists reminders and scheduled tasks in an Upcoming view. Recurring scheduled tasks are limited and mostly arrive as reminders in chat.

Opens files, organizes folders, drafts emails, and builds docs, and can run scheduled recurring tasks. Since September 2026 this runs from any conversation rather than a separate Cowork surface, rolling out on paid plans first.

Its Work agent runs longer tasks across connected apps and files in a hosted browser, including on sites that require you to sign in, and pauses for approval on consequential actions.

Agentic task automation, largely on the top tier and still rolling out.

Its AI browser clicks, fills forms, and runs multi-step tasks across tabs. Being agentic, it asks you to confirm sensitive actions.

Generally available inside Word, Excel, and PowerPoint since April 2026, and strong at acting on your Microsoft 365 content.

A standout: runs autonomous multi-step tasks and coordinates many sub-agents in parallel for research and coding. Behind paid tiers, and still maturing.

Named AI teammates that each work on a persistent cloud computer, sign into sites and apps with your credentials, keep running when your device is closed, and hand work to each other. A separate app from grok.com, still in beta; all of your Bots share one computer, so xAI says not to treat separate Bots as a security boundary.

The core product. It works in a cloud sandbox, in your own Chrome or Edge browser, or on your own desktop, and can be controlled remotely from your phone. Long tasks can fail partway and still use credits.

The core product. It browses, fills forms, books, negotiates and checks out on your behalf in a dedicated cloud machine, and the Mac app can operate desktop apps. It asks before sending or spending.

50+ integrations via an open standard, free on all plans including free.

Connects to a growing set of apps for reading and limited actions.

The best fit if you live in Google: Gmail, Docs, Sheets, Drive, Calendar, Maps, and YouTube all in reach.

The best fit if you live in Microsoft: deep access to Outlook, Teams, Word, Excel, and the rest via Microsoft Graph.

Strong tool-use under the hood and a browser extension (WebBridge) that operates the web for you on paid tiers; a consumer connector suite for email and calendar is less established.

Read and act on Google Workspace, Microsoft 365 (Outlook, OneDrive, SharePoint, Teams), Notion, GitHub, Linear, Slack, Salesforce, and more, plus custom MCP servers. Write access means Grok can draft or send on your behalf, so review permissions.

Connects to Gmail, Google Drive, Notion, Slack, GitHub, Stripe, HubSpot and more, plus custom MCP servers. You can also start tasks by emailing Mail Manus or messaging it in Slack. Every connector is on the free plan, but limited credit caps can make heavy use difficult.

Built-in connectors cover email, calendar, shopping, finance, health and many business tools. Muse can build a Custom Connector for anything else, though Meta doesn't review custom ones.

Relays a task from the Claude mobile app to a local Claude Code session on your desktop, where it runs locally with your files, connectors, and tools, then sends the result back to your phone. A relay, not cloud compute — your machine has to be awake. Launched March 2026.

Ask Manus from your phone to work on your own computer and watch the desktop live while it does. Equivalent to Dispatch in Claude Code.

Reads images, screenshots, diagrams, and PDFs well — but can't take audio or video as input. Text and images only.

Handles images, documents, and audio; strong at reading screenshots and analyzing uploaded files.

Natively understands image, audio, and video, including frame-by-frame video analysis.

Can read uploaded images and files, but it's a search engine first — analysis isn't its focus, and uploads are capped on free.

Reads uploaded images and files, and Vision can watch your screen or phone camera live to talk through what it sees.

Reads uploaded images, screenshots, and documents including long PDFs and spreadsheets, but can't take audio or video as input.

Natively understands images and video alongside text and handles very large documents in one pass — a core strength.

Analyzes uploaded images, PDFs, Office files, audio, and video (MP4/MOV), up to about 100 files per message on the web and 150 MB per file. Transcription quality on audio and video varies.

Reads uploaded documents, spreadsheets and images, and checks screenshots while browsing.

Understands files, photos, audio, and other media you upload or store in Muse, plus content from connected apps.

Toggle deeper step-by-step reasoning for complex analysis, and watch it work through the logic.

Switch to a thinking model for harder reasoning; the deepest tiers are paid.

Explores multiple solution paths in parallel for complex coding, math, and science problems.

Breaks complex queries into sub-questions and can route to reasoning models, but there's no simple visible 'think longer' toggle like the others.

A 'Think Deeper' mode gives more detailed, reasoning-heavy answers versus quick responses.

A toggle that switches into a slower mode showing its full step-by-step reasoning, tuned for math, logic, and tricky code.

A dedicated thinking mode for harder reasoning, math, and code, alongside a faster instant mode.

A mode that thinks step by step before answering; Heavy goes further by running several agents in parallel. Expert requires a paid plan, though Auto may route harder free questions to it.

Max mode spends longer on complex tasks, and Plan Mode shows the plan for you to edit before it runs. There is no separate thinking toggle for chat answers.

Account-level memory carries preferences and working style across sessions when turned on.

Robust, user-configurable memory across chats.

Remembers context and can draw on it in voice chats; less granular control than ChatGPT.

Exists and now feeds voice chats, but narrower and less configurable than ChatGPT's, and doesn't carry full context between sessions the same way.

Records preferences and details as you talk and carries them into later chats; capped at 50 saved entries of 500 characters each, viewable and editable in settings.

Remembers details across chats; you can view and delete individual memories, turn it off under Data Controls, or use Private Chat for conversations it won't keep. Not yet available in the EU and UK, though xAI has said it's coming.

A user knowledge base of saved entries applies across tasks. Manus doesn't automatically learn from past chats.

Remembers details across sessions, including from connected apps, and acts on them unprompted. You can tell it to forget specific things.

Choose among OpenAI models only; naming (Instant / Thinking / Pro variants) is confusing.

Choose among Google's models only.

Unusually, you can route to models from OpenAI, Anthropic, and Google under one subscription.

Switch between a fast everyday model and a slower reasoning mode in one tap, free — but only among DeepSeek's own models.

Switch between instant and thinking modes and among Kimi's own models — no cross-vendor choice.

You pick a speed-versus-depth mode rather than a named model; Auto routes between Fast and Expert. Expert needs a paid plan and Heavy needs the top tier.

Manus picks the model. You choose a lighter, standard or Max mode. Manus Flex lets you bring your own API key and model from OpenRouter, Fireworks or Modal; the provider bills inference and Manus credits still apply.

Spoken conversations in the mobile app.

Fluid, interruptible voice that feels close to a real conversation.

Natural voice in 45+ languages; can also see through your camera or screen while you talk.

Voice mode on desktop and mobile, built on a third-party realtime speech model.

Natural voice in 50+ languages, and it can reference your stored memories.

Real-time spoken conversation with six app voices, adjustable speed, selectable personalities, and a camera option so Grok can discuss what you're pointing at. Available on Free with its own separate limit.

Transcription and conversational voice modes are live as of September 2026.

Surfaces products with prices and reviews, but after pulling back its in-chat checkout in early 2026 it now hands you off to the merchant's own site to buy.

Checks out inside the chat through Google Pay for participating retailers.

Completes the purchase inside the chat via PayPal across thousands of merchants, without leaving the conversation.

A comparable step-by-step tutoring mode, less developed than Gemini's and not tied to classroom tools.

A tutoring mode that walks you through a topic step by step and adapts to your level rather than just handing over the answer. ChatGPT has a comparable study mode, but Gemini's is the most developed and ties into its Classroom tools.

A study mode that works through a topic step by step and builds flashcards and quizzes from your material, rather than returning a finished answer. On Pro and Education Pro.

An on-device mode that captures and summarizes a spoken meeting or conversation. Gemini and Copilot also record meetings, though ChatGPT's is among the more platform-agnostic.

SOC 2 and standard enterprise-plan controls like SSO and data retention, but not a deep governance suite.

Enterprise tier adds SSO, an admin console, and data controls; lighter than Microsoft's governance stack.

Rides Google Workspace admin controls; the standalone enterprise offering is still maturing, with some admin-tooling churn.

An enterprise plan and admin controls exist but are newer and narrower.

Purview data-loss prevention, sensitivity labels, watermarking of AI output, and endpoint DLP, aimed at regulated industries.

Team workspaces with no-training terms, SOC 2, role-based access, SSO, SCIM, audit controls, custom data retention, and customer-managed encryption keys (the last several on Enterprise); a BAA and DPA are published. Grok has faced active regulator scrutiny in the EU and UK in 2026.

Team plans add admin controls and a shared credit pool, with SSO as an add-on. The Trust Center lists SOC 2 Type II and ISO 27001. During the August 2026 Meta separation, some user data was deleted.

DeepSeek publishes its model weights under an open license, so a technically capable team can download and run it on their own hardware for full data control.

Moonshot publishes Kimi's model weights under an open license, so a technically capable team can download and self-host it for full data control.