Skip to content

General

ChatGPT Alternatives: Assistants, APIs and Self-Hosting

ChatGPT alternatives compared by what they are good at: other assistants, the API with your own interface, open models, migration tips and MCP tools.

People look for a ChatGPT alternative for different reasons: a model that writes or codes better for their work, answers with sources, a lower or more predictable cost, more control over their data, or a way to use AI inside their own tools. The right alternative depends on which of those you need, so this guide groups the options by reason rather than ranking them.

It covers four routes. The first is another full assistant with its own app. The second keeps OpenAI's models, or anyone else's, but replaces the ChatGPT app with an interface you control. The third runs open-weight models on your own hardware. The fourth keeps ChatGPT and fixes what it cannot do by giving it tools. After those come a walkthrough for choosing, tips for moving your history and instructions, a comparison table and answers to the questions people ask most.

Why people look beyond ChatGPT

Before comparing products, it helps to name the problem you are trying to solve, because each one points somewhere different.

  • Quality for a specific job. ChatGPT is a generalist. For long-form writing, a large codebase or a stack of contracts, another model may simply suit your work better.
  • Sources and verification. If you spend your time checking whether an answer is true, an assistant built around web search and citations saves more time than a better writer.
  • Cost and seats. A per-seat subscription is predictable for one person and expensive for a team where most people use AI lightly. Pay-per-use can be cheaper, or much more expensive, depending on volume.
  • Data and compliance. Some organisations cannot send certain data to a consumer app at all, or only to providers in a particular region or under a specific contract.
  • Where the work happens. If your documents live in Google Workspace or Microsoft 365, an assistant built into those tools can see context ChatGPT cannot.
  • Capabilities. Sometimes the model is fine but it cannot reach the thing you need: a live DNS record, a real file conversion, an accurate date calculation.

Most people end up with a combination: one main assistant plus an API, a local model or a set of tools for the cases it handles badly.

Other AI assistants

These are full assistants with their own apps, comparable to ChatGPT. All of them offer a free tier with limits and paid plans with higher limits and more features, and business plans with admin controls and different data terms. Plans and limits change often, so check each provider's current pricing page rather than relying on a comparison article.

Claude

Claude (Anthropic) is known for long, careful writing, strong coding help and large documents. It has projects for grouping files and instructions, and it connects to tools through MCP connectors.

Where it is good: Claude tends to follow detailed instructions closely and holds a consistent tone over long pieces of writing, which makes it popular for drafting, editing and reports. It handles long documents well, so you can give it a full specification or a set of papers and ask questions across them. For software work, Anthropic also makes Claude Code, an agent that works in your terminal and repository.

Where it falls short: its built-in feature set is narrower than ChatGPT's in areas such as image generation, and heavy users on consumer plans can run into usage limits during long sessions. Like every assistant, it can be confidently wrong, so facts still need checking.

Who it suits: writers, analysts, developers and anyone whose work involves long documents or detailed instructions. If you are weighing Claude itself against other options, our guide to Claude alternatives looks at the same question from the other side.

Gemini

Gemini (Google) is built into Google's products, from Gmail and Docs to Android, which makes it convenient if your work already lives in Google Workspace.

Where it is good: the integration is the point. Gemini can work with your mail, documents and files where you already keep them, and on Android it can take the place of the old voice assistant. It handles long inputs well and has strong multimodal features for images, audio and video. Gems let you save reusable instructions, similar to custom GPTs.

Where it falls short: outside the Google ecosystem much of that advantage disappears, and some people find its writing less distinctive than Claude's or ChatGPT's. Business features depend on your Workspace edition and its admin settings.

Who it suits: individuals and teams already on Google Workspace or Android. For a closer look, see our Google Gemini alternatives guide.

Microsoft Copilot

Microsoft Copilot is built into Windows, Edge and Microsoft 365, and is the natural choice in organisations standardised on Microsoft.

Where it is good: in Microsoft 365 it works inside Word, Excel, Outlook, Teams and PowerPoint, and in business settings it can draw on your organisation's documents and mail within the permissions you already have. Microsoft has a long history of enterprise compliance work, which often makes procurement easier.

Where it falls short: the consumer Copilot and the Microsoft 365 business Copilot are separate offerings with different features, which confuses many people. Outside Microsoft's apps it is a fairly standard assistant.

Who it suits: companies on Microsoft 365, especially where IT already manages licences and data policies through Microsoft.

Perplexity

Perplexity is an answer engine: it searches the web for each question and cites its sources, which suits research and fact-finding more than open-ended writing.

Where it is good: every answer comes with links, so you can check a claim in seconds. It is fast for questions about current events, product comparisons, documentation and anything where the answer is on the web somewhere. Spaces let you collect research on a topic.

Where it falls short: citations make answers easier to verify, not automatically correct. Summaries can still misread a source, and the quality depends on what the search finds. For long creative writing or deep coding sessions, a general assistant usually does better.

Who it suits: researchers, students, journalists, buyers comparing products and anyone who mostly uses ChatGPT as a better search engine.

Le Chat

Le Chat (Mistral AI) comes from a European company that also publishes open-weight models, which matters to some teams for data residency and vendor choice.

Where it is good: it is quick and capable for everyday tasks, and Mistral's open-weight models mean you can start with the hosted assistant and later run a related model on your own infrastructure. For European organisations, a European provider can simplify legal review.

Where it falls short: its ecosystem of integrations and third-party apps is smaller than those of the largest providers.

Who it suits: European teams, organisations that want vendor diversity, and anyone who likes the idea of a path from a hosted assistant to self-hosted models from the same lab.

Others worth knowing

The field keeps moving. Grok from xAI is built into X, DeepSeek and Qwen come from Chinese labs that also publish strong open-weight models, and many specialised assistants focus on a single job such as meeting notes, writing or customer support. If your organisation has rules about where data may be processed, check the provider's jurisdiction and data terms before trying any of them with real work.

How to test assistants fairly

Try two or three with your own real tasks for a week. The differences show up quickly in practice and rarely in feature lists. Pick five tasks you actually did last month, give each assistant the same prompt and files on the same day, and judge accuracy, editing needed, tone and speed. Free tiers are enough for this.

The API with an interface of your own

If you like OpenAI's models but not the app, or want one interface for several providers, use the APIs directly. You pay per token instead of per seat, and you choose an interface: open-source projects such as LibreChat, Open WebUI and LobeChat connect to OpenAI, Anthropic, Google and others. Our guide to self-hosting a ChatGPT-style interface walks through the setup, and OpenAI alternatives compares other model providers.

How API billing differs from a subscription

A subscription charges a fixed amount per person per month and limits how much you can use. An API charges for what you send and receive, measured in tokens, with different rates per model. Light users often pay less than a subscription; heavy users with long documents and long conversations can pay more. Most providers let you set spending limits and show usage in a dashboard, so set a limit on day one.

Two details catch people out. First, every message in a conversation resends the history, so long chats cost more per message as they grow. Starting a new chat for a new topic keeps costs down. Second, larger models cost much more per token than smaller ones, and for routine tasks a smaller model is often good enough.

What you gain and what you give up

You gain choice: one interface for OpenAI, Anthropic, Google, Mistral and open models, with the ability to switch models mid-conversation or compare answers. You choose where history is stored, and API terms for business use are often different from consumer app terms, so read them for your case.

You give up the polish and the extras of the consumer app. Features such as built-in memory, voice mode, image generation, file handling or web search depend on the interface you pick and the plugins you configure. Someone also has to run and update the interface if you self-host it, and keep the API keys safe.

Gateways and cloud platforms

If you want several providers behind one key, routing services such as OpenRouter offer one API for many models. Organisations that already run on a large cloud can often reach the same models through their cloud account, for example OpenAI models through Microsoft Azure, and Claude through Amazon Bedrock or Google Cloud's Vertex AI. That keeps billing, access control and data agreements in the place your company already manages them.

Open models on your own hardware

When data may not leave your network, run an open-weight model locally. OpenAI's gpt-oss models, Google's Gemma, Meta's Llama, Mistral's open models and others run through tools like Ollama and LM Studio. They are less capable than the largest hosted models, but they are private and work offline. See local AI models for the tools.

What you need

The main constraint is memory. Small models run on an ordinary laptop; larger ones need a graphics card with plenty of video memory or a computer with a lot of unified memory. Quantised versions of a model trade a little quality for a much smaller footprint, which is how most people run models at home. Start with a small model, see whether it is good enough for your task, and only then think about hardware.

Where open models fit

Local models are a good fit for summarising and searching private documents, drafting where the content is confidential, coding help on proprietary code, working on a plane or in places with poor connectivity, and experimenting without a bill. They are a weaker fit for tasks that need the strongest reasoning available or up-to-date knowledge of the world, since a local model knows only what it learned in training unless you connect it to search or tools.

Interfaces for local models

Ollama and LM Studio both run models and expose a local API. Open WebUI gives that API a ChatGPT-like interface in the browser, and LibreChat and LobeChat can mix local and hosted models in one place. Many teams use exactly this setup: a local model for sensitive material and a hosted model for everything else, in the same interface.

Tools for the assistant you already use

Sometimes the real problem is not the chat app but what it can do. An assistant that can only talk will tell you how to check a DNS record or convert a file; one with tools can do it. The Model Context Protocol (MCP) is the open standard for that: an assistant connects to an MCP server and calls its tools during the conversation. ChatGPT, Claude, Cursor and many other clients support it. Our explainer on what an MCP server is covers how it works.

Why tools beat a better model for some tasks

Language models are good at language and unreliable at exact work. They can miscount characters, get time zone arithmetic wrong, invent a DNS record or produce a regular expression that almost works. A tool returns the real answer: it actually performs the lookup, runs the conversion or tests the pattern. If your frustration with ChatGPT is that its answers to precise questions need checking, adding tools usually helps more than switching to another assistant with the same weakness.

MultitaskAI

That is what MultitaskAI is today. It used to be a ChatGPT alternative itself: a browser-based chat interface for your own OpenAI, Anthropic and Google API keys, with split-screen chats and agents, sold as a one-time licence. That app is retired. MultitaskAI is now a remote MCP server that gives AI agents 166 exact tools at one endpoint, so you can keep ChatGPT, or any alternative, and make it more capable.

The tools cover text, web development, images, units, dates and time, domains, SEO, finance and health. In practice that means DNS, WHOIS and SSL lookups, JSON repair, regex testing, unit and format conversions, image resizing and many smaller utilities that a model would otherwise guess at. It works with ChatGPT, Claude, Claude Code, Cursor, VS Code, Windsurf, Gemini CLI, Codex CLI and any MCP client over HTTP. Connectors in ChatGPT and Claude sign in with OAuth; editors and command-line tools use an API key.

Every signed-in account gets 100 free requests a day, which reset at midnight UTC. MultitaskAI Pro removes the limit for €9 a month. The pricing page has the details.

Because MCP is a standard, the same tools follow you if you switch assistants later. That makes tools a safe investment even while you are still deciding which assistant to use.

How to choose: a practical walkthrough

Feature lists rarely settle the question. Working through your situation does. Here are common scenarios and where each one usually leads.

You write for a living

If most of your use is drafting, editing and rewriting, test Claude against ChatGPT with your own material. Pay attention to how much you edit the output, how well each one keeps your voice, and how it handles long documents. Many writers settle on whichever needs less correction for their style, and that varies by person.

You research and need sources

If you mostly ask questions whose answers exist on the web, try Perplexity first. Its citations change the workflow: you read the answer, open the two sources that matter, and move on. Keep a general assistant for writing up what you found.

Your company runs on Google or Microsoft

The integrated assistant usually wins on convenience, because it can see your documents and mail within existing permissions and your IT team can manage it. Start there, check whether it handles your core tasks well, and add a second assistant only for what it does badly.

You are a developer

The comparison is less about chat apps and more about coding tools. AI editors such as Cursor, VS Code with an assistant, Windsurf, and terminal agents such as Claude Code, Codex CLI and Gemini CLI all work inside your project. Most of them can use models from several providers and call MCP tools, so pick the editor or agent you like and treat the model as a setting. Our setup guide for Cursor shows how to add tools there.

You need data to stay private

Decide first what "private" means for you. If it means "not used for training and covered by a contract", a business plan or an API agreement with the right terms may be enough. If it means "never leaves our network", run open models locally or on your own servers, and accept that they are less capable than the largest hosted models.

You use AI lightly, or a team does

For one occasional user, a free tier is often enough. For a team where a few people use AI heavily and most use it lightly, per-seat plans can be wasteful; an API with a shared self-hosted interface lets you pay for actual use. Set spending limits and keep an eye on the first month's bill.

You like ChatGPT but it keeps getting precise things wrong

Do not switch yet. Add tools through MCP and see whether the problems go away. Lookups, conversions, date calculations and validation are exactly the tasks where a tool beats any model.

Moving from ChatGPT without losing your work

Switching assistants is easier than it used to be, but your history, instructions and habits do not move by themselves.

Export your history

ChatGPT lets you export your data from its settings, under data controls. You receive a download link by email, and the archive contains your conversations in a machine-readable file and a readable HTML version. Keep it somewhere safe even if you never import it anywhere. Most other assistants cannot import ChatGPT history directly, but you can search the export, copy useful conversations into documents, or add selected ones as files to a project in your new assistant.

Move your custom instructions and memory

Copy your custom instructions and personalisation settings into a text file before you leave. Then review what ChatGPT has saved in memory and copy anything worth keeping. In the new assistant, put general preferences in its equivalent setting (profile preferences, saved information or a default style) and put project-specific context in a project or a saved assistant. It is a good moment to rewrite them: instructions tend to grow by accretion, and a short, clear set usually works better.

Rebuild custom GPTs

Custom GPTs do not transfer, but their parts do. Each one is mainly instructions, some files and sometimes actions. Copy the instructions and files, then recreate them as a Claude project, a Gemini Gem or a saved assistant in your new interface. Actions that called an API are often easier to replace with an MCP server, which works across clients rather than inside one.

Keep prompts outside any assistant

Store prompts you reuse in a notes app or a folder of text files, so the next switch costs minutes rather than days.

Run both for a while

Keep your ChatGPT account for a month after you move. Some tasks will turn out to work better in the old tool, and you will want access to past conversations while the new habits settle.

Comparison at a glance

Option Best at Main trade-off Good fit for
Claude Long-form writing, coding, large documents Fewer built-in extras such as image generation Writers, analysts, developers
Gemini Google Workspace and Android integration, long inputs Less advantage outside Google's tools Google-based teams
Microsoft Copilot Microsoft 365, Windows and Edge integration Consumer and business versions differ Microsoft-based organisations
Perplexity Web answers with cited sources Weaker for long writing and coding Research and fact-finding
Le Chat European provider with open models Smaller integration ecosystem European teams, vendor diversity
API with your own interface Pay per use, several models in one place You run the interface and manage keys Teams with uneven usage, tinkerers
Open models locally Privacy and offline use Less capable than the largest models, needs hardware Sensitive data, offline work
MCP tools such as MultitaskAI Exact lookups, conversions and checks Adds abilities, not a new model Anyone who wants their current assistant to do more

And the short version, by goal:

If you want Look at
Better long-form writing or coding help Claude
AI inside Google Workspace or Android Gemini
AI inside Microsoft 365 and Windows Microsoft Copilot
Answers with cited sources Perplexity
A European provider with open models Le Chat
Pay per use and one interface for several models The APIs with a self-hosted interface
No data leaving your network Open models run locally
Your current assistant doing more MCP tools such as MultitaskAI

Frequently asked questions

What is the best free ChatGPT alternative?

There is no single answer: Claude, Gemini, Copilot, Perplexity and Le Chat all have free tiers with different limits. Test them on the tasks you actually do. If you need privacy rather than a free app, an open model running locally costs nothing beyond your hardware.

Can I use ChatGPT's models without ChatGPT?

Yes, through the OpenAI API with your own key and an interface of your choice. It is billed separately from ChatGPT plans, so a ChatGPT subscription does not include API usage and an API key does not unlock the ChatGPT app.

Is there a ChatGPT alternative that runs offline?

Yes. Open-weight models run locally with tools like Ollama or LM Studio, with no internet connection once the model is downloaded. Expect them to be slower and less capable than the largest hosted models, particularly on modest hardware.

Is the API cheaper than a ChatGPT subscription?

It depends on how much you use it. Occasional users often pay less through the API, while people who paste long documents and keep long conversations going can pay more. Set a spending limit, use smaller models for routine tasks and check your usage after the first few weeks.

Can I take my ChatGPT history to another assistant?

You can export it from ChatGPT's data controls, but most assistants cannot import it directly. The practical approach is to keep the export as an archive, copy the conversations you still need into documents, and add those to a project in the new assistant.

Which alternative is best for coding?

For serious coding, look at AI editors and terminal agents rather than chat apps: Cursor, VS Code, Windsurf, Claude Code, Codex CLI and Gemini CLI. Most of them let you choose the model, so try a couple of models inside the tool you prefer.

Do I have to switch assistants to get tools like DNS lookups or file conversions?

No. ChatGPT supports MCP, so you can connect a tool server and keep using it. The same server also works with Claude, Cursor and other clients, so the tools stay with you if you switch later.

Is MultitaskAI still a chat app?

No. The chat app is retired. MultitaskAI is now an MCP server with 433 tools for agents, and works with ChatGPT and other MCP clients. Every signed-in account gets 100 free requests a day, and the setup guides show how to connect it.

Give your agent 433 tools in a minute

Connect MultitaskAI to your MCP client and ask in plain words. 100 requests a day are free.