Weighing ChatGPT vs Claude vs Gemini, or Copilot and Grok? This lesson sets out who makes each one, what each offers a business in its maker’s own words, and how to test them on your own work. It does not name a winner, because the honest answer depends on the task.
In this lesson
- The handful of companies that make the models underneath most AI products.
- The four ways a business gets AI, and when each fits.
- What a business plan changes, as questions to ask any vendor.
- What open-weight models are, and why some firms run them on their own computers.
- Why “which task?” matters more than “which model?”, and how to test models on your own work.
Model makers and products
There are thousands of AI products, but only a handful of companies make the large models most of them run on. A model maker trains the model (Lesson 1). A product is what you actually use: a chat app, a feature inside your email, a phone answering service. Some makers sell their own products; many products are built on another company’s model.
The names can be confusing because the same word is often used for the company’s assistant, its models and its business plans. Roughly:
- OpenAI makes the GPT models and the ChatGPT assistant.
- Anthropic makes the Claude models and the Claude assistant.
- Google makes the Gemini models, the Gemini assistant, and Gemini features inside Google Workspace.
- Microsoft sells Microsoft Copilot (until recently “Microsoft 365 Copilot”) inside Word, Excel, Outlook and Teams. Microsoft says it offers models from other makers in Copilot, including Anthropic’s and OpenAI’s, that administrators choose whether to use them, and that it may also run models it hosts itself (Microsoft Learn).
- xAI (whose pages now carry the name SpaceXAI) makes the Grok models and the Grok assistant, also built into X.
- Meta makes Meta AI and publishes some of its models for anyone to run.
- Mistral AI makes its own models and an assistant now called Mistral Vibe (formerly Le Chat).
- DeepSeek and Alibaba (with its Qwen models) are among the makers that publish many of their models openly.
The table further down sets out each maker’s assistant, business plans, agent features, data terms and prices, as each lists them.
Four ways a business gets AI
A ladder of four steps. Each step takes more effort, gives more control and costs more. 1: a chat assistant in a browser. 2: a business plan of that assistant. 3: AI built into software you already pay for. 4: building on a model through its API.
- 1Chat assistant in a browserPaste text in, copy answers out. Personal or free plan.
- Effort
- Control
- Cost
- 2Business plan of that assistantAdmin controls, shared workspaces and the vendor's business data terms.
- Effort
- Control
- Cost
- 3AI built into software you already pay forAssistants inside your email, office suite, accounting or booking software.
- Effort
- Control
- Cost
- 4Build on a model through its APIA developer connects a model to your own systems. You pay per use.
- Effort
- Control
- Cost
- A chat assistant in a browser. Anyone can sign up to a free or personal plan of ChatGPT, Claude, Gemini, Copilot, Grok or others, and start typing. Quick and cheap, but on personal plans you have little control over how your data is handled, and nothing is shared or managed across your team.
- A business plan of that assistant. The same assistant on a plan for organizations, priced per user per month. You get admin controls, shared workspaces and, most importantly, stated terms on how your business data is used. For most small businesses this is the right first step.
- AI built into software you already pay for. Your email, office suite, accounting, help desk, booking or design software may already include AI features, or offer them as an add-on. Often the easiest route, because the AI already sees the data it needs. Check what you have before buying anything new.
- Building on a model through its API. A developer (yours or a supplier’s) connects your own systems to a model directly and pays per use. This gives the most control and is how custom bots and agents are built, but it needs technical skill and ongoing care.
As you go down the list, effort, control and usually cost rise. Most businesses use routes 2 and 3, and reach route 4 only through a supplier.
What a business plan changes
Business plans differ from free ones mainly on data handling, administration and limits. Do not assume: read the vendor’s own words. Ask these questions of any plan:
- Is our data used to train the vendor’s models? Many business plans say no by default, but not all do. For example, Mistral’s help centre says that on its Vibe plans other than Enterprise, “input and output data are used by default to train our artificial intelligence models” unless you opt out, while “Enterprise customers are opted out of training by default” (Mistral AI). The table below quotes each maker.
- How long is our data kept, and can we delete it?
- Where is it stored? This matters if you have customers in the EU or the UK, or work under industry rules.
- What admin controls are there? Adding and removing users, controlling which features and connectors staff can use, seeing usage.
- What are the limits? Many plans cap use of the most capable models or features, even when the plan says “unlimited”.
- What happens if we leave? Can you export your conversations, files and settings?
ChatGPT vs Claude vs Gemini, and the rest: what each maker offers
One row per maker, in alphabetical order. Facts and quotes come from each vendor’s own pages, with the date we read them. We have not tested these products, and list order is not a ranking. For the ChatGPT vs Claude vs Gemini question, read the OpenAI, Anthropic and Google rows side by side: business plans, where each is built in, agent features, what each says about training on your data, and the price as listed. Microsoft’s row covers Copilot and xAI’s covers Grok.
| Maker | Assistant | Business plans | Built into | Agent features | Data use (business, in the maker’s words) | Price as listed | Date read |
|---|---|---|---|---|---|---|---|
| Alibaba (Qwen) | Qwen Chat | Alibaba Cloud Model Studio (pay-as-you-go model API) | Not stated on the pages we read. | Not stated on the pages we read. |
| October 6, 2026 | |
| Anthropic | Claude |
| Anthropic's own add-ins for other makers' software, as listed on the Claude pricing page: Claude in Chrome, Claude for Microsoft 365 and Claude for Microsoft Outlook (Pro, Max, Team and Enterprise). Microsoft also offers Anthropic models inside Microsoft Copilot (see the Microsoft row). | Included in Claude plans: Pro $17/month with annual billing ($200 billed up front) or $20 billed monthly; Max 5x $100/month; Max 20x $200/month. Team: Standard seat $20/seat/month billed annually ($25 monthly); Premium seat $100/seat/month billed annually ($125 monthly). Enterprise: US$20/seat/month billed annually plus usage at API rates. Not included in the Free plan. As listed on October 3, 2026. | October 6, 2026 | ||
| DeepSeek | DeepSeek (chat at chat.deepseek.com) | DeepSeek API (pay per token from a topped-up balance) | DeepSeek Harness | Not stated on the pages we read. |
| October 6, 2026 | |
| Gemini (Gemini app; Gemini Enterprise app for business) |
| Google Workspace: Gemini in Gmail, Docs, Sheets, Meet and more, the Gemini app and Gemini Notebook are listed on the Workspace plans; the privacy hub also covers Gemini in Chrome. |
| October 6, 2026 | |||
| Meta | Meta AI |
| Meta Business Agent starts with the WhatsApp Business app, Instagram Pro, Messenger and Meta Business Suite; the Meta AI assistant runs on AI glasses and the Meta AI app and web, per Meta's connectors page. | Not used to improve our products.Source |
| October 6, 2026 | |
| Microsoft | Microsoft Copilot (formerly Microsoft 365 Copilot); Microsoft Copilot Chat (formerly Microsoft 365 Copilot Chat) |
| Microsoft 365 apps: Copilot in Word, Excel, PowerPoint, Outlook, OneNote and Microsoft Teams; Copilot Mode in Edge; the Microsoft Copilot app. Also Security Copilot, GitHub Copilot and LinkedIn Hiring Assistant (see tool records). | Prompts, responses, and data accessed through Microsoft Graph aren't used to train foundation LLMsSource |
| October 6, 2026 | |
| Mistral AI | Mistral Vibe (formerly Le Chat) |
|
| October 6, 2026 | |||
| OpenAI | ChatGPT |
| Listed on chatgpt.com: ChatGPT for PowerPoint and ChatGPT in Slack and Teams; Microsoft lists 'Latest OpenAI models: GPT-5.X' in Microsoft Copilot (see the Microsoft row). | Your data is never used for training our models.Source |
| October 6, 2026 | |
| xAI | Grok |
| Grok on X (X is a separate company, per SpaceXAI); SpaceXAI's own add-ins for Microsoft Word, Excel, PowerPoint and Outlook and an add-on for Google Workspace; Grok models listed in GitHub Copilot, Microsoft Foundry, Amazon Bedrock and Gemini Enterprise Agent Platform (news page). |
| October 6, 2026 |
Prices and features change often, sometimes monthly. Always check the vendor’s own page before you buy.
Open-weight models
Most of the assistants above run on models that only their maker can run. Some makers also publish open-weight models: the trained model files themselves, which anyone can download and run on their own computers or a cloud server they control. Meta, Mistral AI, DeepSeek and Alibaba’s Qwen team are among the makers that publish models this way, each under its own licence; the table notes which.
Why would a business run one?
- Data stays in-house. Nothing leaves your own systems, which can matter for confidential or regulated information.
- No per-use fee to a model maker, once the hardware is paid for.
- Control. You choose the version and it does not change under you.
The trade-offs: you need technical skill to set it up and keep it secure, and suitable hardware; the most capable models are not always available as open weights; and you take on the work the vendor would otherwise do. Each model’s licence also sets what you may do with it commercially, so read it. For most small businesses, open-weight models arrive through a supplier’s product rather than being run in-house.
Which AI is best? Start with which task
People spend a lot of time asking which assistant is “the best”. It is the wrong first question, for three reasons.
- The leaders change month to month. A model that leads one test in spring is often overtaken by summer.
- Different models do better at different things, and the differences on everyday business tasks are often smaller than the differences in how well the task is set up.
- What matters is your work. A model that tops a public test may do worse on your price list and your customers’ questions.
How models are tested
Independent and published evaluations are useful for understanding the field, as long as you know what they measure:
- Head-to-head grading by experts. In OpenAI’s GDPval, experts with “an average of 14 years of experience” compared AI models’ work with professionals’ work on real tasks from 44 occupations, without being told which was which (Patwardhan et al., 2025). Lesson 3 covers what it found.
- Public leaderboards built from votes. On Arena (formerly LMArena), “you’ll be served 2 anonymous models,” you pick the answer you prefer, and the votes build public leaderboards; “after voting, the model identities are revealed” (Arena). It measures which answers people prefer, not whether they are correct.
- Coding and reasoning benchmarks. Fixed sets of problems with right answers, run on each new model. Useful for comparing raw ability, far from most business tasks.
We name these tests, not a winner. Any “leader” would be out of date by the time you read this.
Test on your own work
The test that counts is the one you run yourself. It takes an afternoon:
- Pick three real tasks from your business: say, a customer reply, a summary of a long document, and a piece of analysis from a spreadsheet.
- Run each through two or three assistants, on business plans or free trials, giving each exactly the same instructions and material.
- Have the person who normally does the task grade the results blind: hide which assistant produced which.
- Note how much checking and fixing each needed, not just which read best.
- Repeat in six months; the answer may change.
This is what the large independent tests do at scale, on your own work instead of someone else’s.
Key terms
- ModelThe AI program underneath an assistant. It is trained once on huge amounts of material, then used to answer questions.One product may offer several models, from fast and cheap to slower and more careful.Lesson 1: What AI is · More in the glossary
- Large language model (LLM)A model trained on vast amounts of text to predict likely next words. ChatGPT, Claude, Gemini, Copilot and Grok are built on them.It predicts likely text rather than looking facts up, which is why it is fluent and sometimes confidently wrong.Lesson 1: What AI is · More in the glossary
- Business plan (AI)A paid plan of an AI assistant for organizations, with admin controls and stated terms on how your data is handled.Plans differ. Read the vendor’s own words on data use, retention and where data is stored before you buy.Lesson 2: what to look for · More in the glossary
- APIA way for software to use a model directly, without a chat window. Developers build products on it and usually pay per use.More in the glossary
- Open-weight modelA model whose trained files the maker publishes, so a business can run it on its own computers instead of the maker’s service.It keeps data in-house and avoids per-use fees, but needs technical skill and hardware to run well.Lesson 2: Who makes AI · More in the glossary
- Training dataThe material a model learned from before release. It knows nothing after its cut-off date, or about your business, unless given it.More in the glossary
- GDPvalA test by OpenAI in which experts compared AI’s work with professionals’ work on real tasks from 44 jobs, without knowing which was which.It is run by an AI company, so we weight it as a vendor study.More in the glossary
Every term is explained in the glossary.
Try this
Find out what AI you already pay for. List the software your business subscribes to (email and office suite, accounting, help desk, booking, design, phone system) and check each vendor’s site or your admin panel for AI features included in your plan or offered as an add-on. Then pick the one assistant your team uses most, find its data-use terms on the vendor’s own site, and write down in one line whether it uses your business data for training. If you cannot find the answer, that is your answer: do not put business data into it until you can.
Sources
- Vendor pages for Alibaba (Qwen), Anthropic, DeepSeek, Google, Meta, Microsoft, Mistral AI, OpenAI and xAI, as listed row by row in our vendor data file (pipeline/data/learn/vendors-2026-10.json), fetched 6 October 2026. Prices for Anthropic and OpenAI as recorded in our AI tools data, fetched 3 October 2026.
- Microsoft Learn, “Data, Privacy, and Security for Microsoft Copilot”, updated 18 August 2026. Read 6 October 2026. https://learn.microsoft.com/en-us/microsoft-365/copilot/microsoft-365-copilot-privacy
- Mistral AI help centre, “Do you use my user data to train your artificial intelligence models?”. Read 6 October 2026. https://help.mistral.ai/en/articles/347617-do-you-use-my-user-data-to-train-your-artificial-intelligence-models
- Patwardhan, T. et al., “GDPval: Evaluating AI Model Performance on Real-World Economically Valuable Tasks”, OpenAI, arXiv 2510.04374, 5 October 2025. Read 6 October 2026. https://arxiv.org/abs/2510.04374
- Arena (formerly LMArena), “How it works”. Read 6 October 2026. https://arena.ai/how-it-works