ChatGPT Alternatives: Claude, Gemini, Copilot and Others Compared

The main alternatives to ChatGPT are Claude from Anthropic, Gemini from Google, Microsoft Copilot, Perplexity, Meta AI, and open-weight models such as OpenAI’s gpt-oss and Google’s Gemma that you run on your own hardware. None of them is the best at everything, and the useful question is not which one is smartest but which one fits the job. If your work lives in Gmail and Google Docs, Gemini is already there. If it lives in Outlook, Teams and SharePoint, Microsoft Copilot is. If you need answers with sources you can click, Perplexity is built around that. Claude sits across both office suites through connectors and adds a coding agent. What follows compares them on the things that decide it in practice, from each vendor’s own pages as of September 30, 2026.

What to compare in any AI similar to ChatGPT

• Where your files already are. An assistant inside the suite you use reads your mail and documents without uploads; one outside it needs a connector.
• What it can connect to. Built-in connectors, and whether you can add your own tools over MCP, the open protocol most of these now support.
• What happens to your data. Whether chats can train the vendor’s models, what the switch is called, and whether a work account changes the answer.
• Who else uses it. A shared workspace for a team is a different product from a personal chat.
• Whether you need sources. Some jobs need a link behind every claim; others need a good draft.

Benchmarks are left out on purpose. They change with every model release, and a model that wins a test can still be the wrong tool if it cannot see your files. What sits underneath all of these, a large language model, is explained in what is an LLM.

Claude

Claude is Anthropic’s assistant, on the web, desktop and phone, with consumer plans named Free, Pro and Max and business plans named Team and Enterprise. Projects keep files, instructions and chats for one piece of work together; how they compare with ChatGPT’s is covered in Claude Projects vs ChatGPT Projects. Connectors reach Google Drive, Microsoft 365 and other services, and custom connectors to any remote MCP server work on every plan, with Free accounts limited to one. For software work there is Claude Code, a coding agent that works in your repository; see what is Claude Code.

On data, Anthropic’s privacy center says that on the consumer plans it uses your chats and coding sessions to improve its models only if you choose to allow it↗ (opens in a new tab), and that inputs and outputs from its commercial products are not used for training by default.

Gemini

Gemini is Google’s assistant, in the Gemini app and inside Google Workspace: Gmail, Docs, Drive, Meet and Chat. Its reusable instructions are Gems, which Google is turning into skills on personal accounts; see Gemini Gems and Claude Projects vs Gemini Gems. Its research notebooks, formerly NotebookLM, are now Gemini Notebook.

The personal account’s data switch is Keep Activity. Google’s Gemini Apps Privacy Hub↗ (opens in a new tab) says that with it off, future chats are not used to train its AI models unless you send feedback, and are kept with your account for 72 hours; with it on, a subset of chats is read by human reviewers after being disconnected from your account. Work accounts are different: Google’s Workspace privacy hub says your content is not human reviewed or used for model training outside your domain without permission.

Gemini can also take your own tools. Google’s page on custom apps↗ (opens in a new tab) lets you add an MCP server by URL, but for now only on a personal account, for people 18 or over in the US, with Keep Activity on, in the Gemini web or mobile app, in English. A work account cannot add one there.

Microsoft Copilot

Start with the names, because Microsoft keeps changing them. Its page on enterprise data protection↗ (opens in a new tab) says Microsoft 365 Copilot is now named Microsoft Copilot and Microsoft 365 Copilot Chat is now Microsoft Copilot Chat, with some screens and licenses still using the old names. There is one Copilot app, and you sign in with either a personal Microsoft account or a work or school account, switching between them without the data mixing.

With a work account, Copilot answers from your organization’s Microsoft 365 data through Microsoft Graph, respects existing permissions and sensitivity labels, and Microsoft says prompts, responses and Graph data are not used to train foundation models. The same page notes that Copilot can run on Anthropic models, which shows how loosely the app and the model are now tied. For grounded answers from your own files, see Copilot Notebooks vs ChatGPT Projects; for agents, including MCP tools in Copilot Studio, see Microsoft 365 Copilot agents.

With a personal account, Copilot connects to OneDrive, Outlook.com, Google Drive, Gmail, Google Calendar and Contacts, and Microsoft says it does not train on data from connected services. Your conversations can be used for training unless you opt out; Microsoft excludes signed-out users, users under 18 and work accounts. Microsoft retired Podcasts, Deep Research and Group Chat on personal accounts on August 18, 2026, and on September 25, 2026 announced a new Copilot with Home, Code and Autopilot, starting in its Frontier preview program.

Perplexity

Perplexity is built for research: its own documentation describes web-grounded answers with built-in citations, and its developer platform also offers models from OpenAI, Anthropic, Google and xAI. That makes it the natural pick when every claim needs a source you can open. It also runs an official MCP server, so Claude, Cursor or another agent can call its search and research tools; see Perplexity MCP. Check the data settings in your account, and your company’s terms if you use Perplexity at work, before pasting in client material.

Meta AI

Meta AI is the assistant in the Meta AI app and at meta.ai, which Meta is extending across Instagram, Facebook, Messenger, WhatsApp and its glasses. Since April 2026 it runs on Muse Spark↗ (opens in a new tab), a model Meta has not released openly, though it says it hopes to open-source future versions; the new features started in the US. In May 2026 Meta announced Incognito Chat for WhatsApp and the Meta AI app, rolling out over the following months, where conversations are processed in an environment Meta says it cannot access and are not saved. Meta’s announcements center on personal use inside its apps rather than on work files or team workspaces.

Open-weight models you run yourself

An open-weight model is one whose trained weights you can download and run on your own machine or server. OpenAI describes gpt-oss-20b↗ (opens in a new tab) as an open-weight model for low-latency, local or specialized uses, under the Apache 2.0 license, and Google DeepMind describes Gemma 4 as open models that run from cloud servers down to laptops and phones. Tools such as Ollama run them locally.

The gain is control: nothing leaves your hardware unless you send it. The cost is that you host, update and secure it yourself, and you get a model, not an app with projects and connectors. How that works with coding agents is in Claude Code with local models, and why a local model is not the same as a private workflow is in MCP with Ollama.

Alternatives to ChatGPT, by job

• Writing and long documents: Claude or ChatGPT, both with projects for standing instructions and reference files. How to prompt either well is in prompt engineering.
• Research with sources: Perplexity, or a notebook grounded in your own files, Gemini Notebook or Copilot Notebooks; see NotebookLM alternatives.
• Coding: Claude Code or Gemini CLI as agents in your repository; in the chat apps, see Claude vs ChatGPT for coding.
• Work in Google Workspace: Gemini, already in Gmail, Docs and Drive.
• Work in Microsoft 365: Microsoft Copilot with a work account. Claude also reaches Outlook, Teams and SharePoint; see Claude with Microsoft 365.
• Team projects: shared projects in Claude on Team and Enterprise, shared projects in ChatGPT, or shared Copilot Notebooks; the sharing rules differ, and the comparisons above cover them.
• Connecting your own tools over MCP: Claude on every plan, ChatGPT in developer mode, Gemini’s custom apps on eligible personal accounts, and Copilot Studio at work.
• Keeping data on your own hardware: an open-weight model.

Data controls, side by side

• Claude: consumer chats train models only if you allow it; commercial plans are not used for training by default.
• Gemini: Keep Activity decides whether personal chats are saved and used for training; Workspace content stays out of training outside your domain without permission.
• Microsoft Copilot: personal conversations can train models unless you opt out; work accounts are excluded.
• ChatGPT, for comparison: OpenAI’s help center says personal accounts can turn off “Improve the model for everyone”, and business workspaces are not used for training by default.
• Open-weight: whatever your own setup does.

The pattern is consistent: a work account usually changes the default. If you are choosing for a team, choose the account type before the assistant, and agree the rules first; how to use AI at work has a one-page policy to copy.

Keep the work list outside any one assistant

Teams rarely end up with one assistant. Someone drafts in Claude, someone else lives in Copilot, and a developer runs an agent in the terminal. If the list of what needs doing lives in one vendor’s chat history, it is stranded there when you switch; moving between them is a rebuild, as moving from ChatGPT to Claude shows.

fenbs is a board that sits outside all of them. Each task is a feature, enhancement or bug in To Do, Next Up, In Progress or Completed, and Claude and ChatGPT connect over MCP as members with a role, each change recorded in History under who made it. An assistant that cannot connect can still work from the board: Copy as Markdown copies it for pasting into a chat, and Add many turns a pasted list back into tasks, one per line, starting with bug:, feature: or enhancement:. fenbs has no model of its own, and no due dates, settable assignee or sprints, so dates and owners go in the task’s note.