Claude, ChatGPT, Gemini: how to choose

Module 1 separated the model (the engine) from the product (the app). Module 2 showed that almost everything depends on the context you give it. Now for the question people ask first and should ask third: which tool? This module answers it honestly: we use Claude as the example because it is what we use, but the criteria work for any of them.

Everything here about plans, prices and features is as of October 2026 and changes often. What does not change is the logic: what you are buying when you pay, why there are several models inside one app, which built-in tools expand what it can do, and how to decide without hopping to a new app every month.

What are the main AI tools today?

For office work, three general AI assistants account for almost all the usage, and a fourth matters if your company lives in Microsoft:

• Claude, by Anthropic. The app runs the Claude models (Fable, Opus, Sonnet, Haiku). It is on web, desktop and mobile, and it is the basis for this course.
• ChatGPT, by OpenAI. The most widely used app; inside it run models from the GPT family. It has personal, team and enterprise plans.
• Gemini, by Google. The name of both the app and the models (with Pro and Flash variants). Its natural edge is integration with Gmail, Drive and Docs, and the plans are sold as "Google AI".
• Microsoft Copilot. The AI inside Word, Excel, Outlook and Teams. For a company already paying for Microsoft 365, it usually comes in as an add-on to that subscription.
• There are others (Grok, Perplexity), but for a small business the real decision is among these four.

What do I get by paying: free, personal or a team plan?

All four vendors follow the same ladder, and the personal rung costs about 20 dollars a month. What changes at each level:

1. 1Free. Fine for learning and one-off tasks. It usually has message limits, smaller models and fewer features. On Claude, the free plan as of October 2026 includes chat on web, desktop and mobile, web search, file creation and code execution, memory across conversations and connecting your apps.
2. 2Paid personal (about USD 20 a month). You are mostly buying more usage and access to the big models. Claude Pro costs USD 20 a month (USD 17 billed annually) and adds more usage, unlimited projects (the free plan allows only a few), more models, scheduled tasks and Claude Code. ChatGPT Plus costs USD 20 a month according to OpenAI’s help center. Google AI Pro, which includes Gemini, costs USD 19.99 a month. Microsoft 365 Premium, with Copilot, costs USD 19.99 a month. The first three also have a rung at about USD 100 a month or more for heavy use (Claude Max, ChatGPT Pro, Google AI Ultra).
3. 3Team or enterprise (per seat). Here you stop buying capacity and start buying control: managed accounts, central billing, rules about which connectors are allowed, activity logs and, most importantly, different data terms (more on that at the end). Claude Team costs between USD 20 and 25 per seat per month depending on billing, and the team plans from ChatGPT, Gemini and Copilot sit in a similar range. Enterprise plans are negotiated with the vendor and add audit logs, configurable retention and role-based controls.

Which model should I pick inside the app?

This is the decision almost nobody makes and the one that changes results the most. Inside every app there are several models, and the logic is the same across vendors: a big, slow, expensive one that reasons best; a middle one; and a small, fast, cheap one for simple or high-volume tasks. For Claude, as of October 2026, Anthropic describes its lineup like this:

1. 1Claude Fable 5.1: for demanding reasoning and long-horizon agentic work. The most capable, and the slowest.
2. 2Claude Opus 5.5: for long-running agentic coding and knowledge work. Anthropic recommends it as the starting point for most workloads.
3. 3Claude Sonnet 5.5: the best combination of speed and intelligence. Fast, for everyday work.
4. 4Claude Haiku 5.5: for high-volume, latency-sensitive tasks such as classification, extraction and routing. The fastest and cheapest.

All current Claude models read text and images and share the same context window, so the difference is not how much they can read but how hard they think before answering and how long they take. Anthropic suggests two paths: start with the economical model and move up only when it falls short, or start with the capable one and move down once the task is well defined. For a small business a simple rule works:

• Move up a model when the task has many steps, needs judgment (analyzing a contract, planning a negotiation, checking figures) or when the middle model’s answer does not convince you. Before deciding that "AI can’t do this", try the same prompt on the most capable model.
• Move down a model when you repeat a simple task many times: summarizing emails, classifying reviews, translating short messages. The small model answers just as well there and uses far less of your usage limit.
• Don’t switch models to fix missing context. If the answer is generic, the problem is almost always module 2, not the model.

In ChatGPT and Gemini the logic is identical, though the names change every few months. What is worth learning is where the model picker sits in your app and what each level means, not memorizing names.

Which built-in tools change what the assistant can do?

In module 1 we said a model on its own does not search, does not calculate and knows nothing after its knowledge cutoff. The apps solve that with built-in tools. These are the ones that weigh most in a comparison, with what Claude offers as of October 2026 as the reference:

• Web search. The assistant checks the internet before answering and cites sources. In Claude it is turned on from the "+" menu in the conversation and is available on every plan. It is the first line of defense against hallucination on current-events questions.
• Files and images. You can upload PDF, Word, Excel, CSV, text and images (JPEG, PNG, GIF, WebP) and ask about them. Claude analyzes charts and tables inside PDFs of up to 100 pages; on longer documents it reads text only. A model that understands images as well as text is what is called multimodal.
• Code execution and data analysis. The assistant writes and runs a program in an isolated environment to calculate, clean a spreadsheet or draw a chart. It is the answer to the arithmetic problem from module 1: instead of "adding in its head", it adds with code. Claude can also create Excel, Word, PowerPoint and PDF files from the conversation.
• Deep research. A mode in which the assistant plans, runs dozens of searches, reads and delivers a cited report; it takes minutes, not seconds. In Claude it is called Research and lives on the paid plans. Anthropic recommends it when the question needs five or more lookups; for direct questions, web search is enough.
• Voice. Talking instead of typing, useful in the car or for thinking out loud. In Claude it is on every plan on mobile, desktop and web; support for languages other than English is in beta, and you set the language in settings.
• Connectors. The assistant reads your Gmail, Drive, calendar or Slack with your permission, instead of you pasting everything by hand. In Claude, Gmail and Drive are available to all users, and on team plans the admin decides which ones are enabled. Together with projects, that is module 6.

All four vendors offer versions of these tools, under different names and limits. When comparing, do not ask "does it have web search?" but "how well does it cite Spanish-language sources?" and "which plan is it on?".

How do I choose a main tool for myself and my team?

This is delegation from the 4D framework: before choosing the tool, be clear about the work you will hand it. Then evaluate on five criteria, in this order:

1. 1The work you actually do. List the three tasks from your week map (module 1) and test each tool on those, not on demos. Drafting contracts, analyzing spreadsheets and answering customers on WhatsApp reward different things.
2. 2Quality in Spanish. The big models all write Spanish well, but not equally. Watch the register (does it sound Colombian or translated), how it handles accents and local terms (NIT, IVA, propiedad horizontal) and whether the Spanish-language sources it cites are real.
3. 3The tools you already use. If your company lives in Google Workspace, Gemini and Claude’s Google connectors are a point in favor; if it lives in Microsoft 365, Copilot has an integration advantage that is hard to ignore. The best AI in the world loses if your team has to copy and paste everything.
4. 4Admin and data controls. For a team: can you create accounts, remove access, decide which connectors are allowed and see who uses what? What does the contract say about training on your data? This outweighs any new feature.
5. 5Price per seat, in pesos, per year. Multiply seats by months and convert to pesos with taxes. Set it against the hours those three tasks consume today. If the number does not work, the problem is not the price but the task you picked.

With that, choose one main tool and one backup, and resist the urge to bounce between five. The cost of tool-hopping is not the money: it is that you never accumulate projects, memory, a prompt library or habits in any of them. The backup is for cross-checking important answers and for not being stuck if the main one fails. A quick guide:

• You work alone and write, analyze and think with AI all day: main tool Claude or ChatGPT on a personal plan; backup, the other one on free.
• Your team lives in Gmail, Drive and Docs: main tool Claude with Google connectors, or Gemini with Google AI; backup, the other.
• Your team lives in Outlook, Excel and Teams: evaluate Copilot inside Microsoft 365 as the main tool; backup, Claude or ChatGPT for demanding writing and analysis.
• You will build automations or software (parts 3 and 4 of the course): Claude, because of Claude Code and the API; ChatGPT and Gemini have equivalents you can evaluate.
• You handle customer data or confidential information: any of the four, but on a team or enterprise plan, never on a mix of personal accounts.

I want to choose a main AI tool (and a backup) for my team. Before recommending anything, ask me questions, two or three per turn, until you are clear on:

1. How many people we are, what roles they have and how many would use AI each week.
2. The three to five concrete tasks we want help with and how long they take us today.
3. The software we already use (email, documents, internal chat, accounting, CRM) and whether it is Google Workspace, Microsoft 365 or something else.
4. Whether we handle customers’ personal data or confidential information, and who would administer the accounts.
5. The monthly budget per person, in [CURRENCY], and whether we prefer monthly or annual billing.

Once you have that, recommend one main tool and one backup among Claude, ChatGPT, Gemini and Microsoft Copilot. For each, explain: why it fits our tasks, which plan we need (personal, team or enterprise) and why, two advantages and two disadvantages against the others, and what we should test for a month before paying annually.

Initial context: [DESCRIBE YOUR BUSINESS IN TWO LINES]. Do not invent prices: if you are not sure of a current price, say so and ask me to verify it on the official page.

How do I compare two tools (or two models) fairly?

Most comparisons going around are unfair: different question, different context, different model, different day. For one to count, hold everything fixed except the tool: the same prompt (with the context from module 2), the same file, the same day, and the same tools turned on (web search on or off in both). Then grade with a rubric you wrote before seeing the answers, not after. That is discernment with a method.

• Accuracy. Are the facts, figures and names correct? One factual error outweighs ten stylistic ones. Verify against a source, not against your gut.
• Compliance. Did it do what you asked, in the format and length requested, without adding what you did not ask for?
• Spanish. Does it sound like someone at your company or like a translation? Did it use the right terms?
• Sources. If search was on, do the sources exist and say what the answer claims?
• Remaining effort. How many minutes of editing before you can use it? That is the measure that really matters.

Act as an impartial evaluator. I will give you a task and two answers (A and B) produced by two different AI tools from the same prompt. You do not know which tool produced which, and you must not guess.

Grade each answer from 1 to 5 on these criteria and justify each score in one line:
1. Accuracy: facts, figures and names are correct (flag any claim that should be verified).
2. Compliance: it did exactly what was asked, in the requested format and length.
3. Spanish: natural, in the right register for [TYPE OF READER], with no translated feel.
4. Sources: if it cites, are they verifiable and do they support what it says?
5. Remaining effort: how many minutes of editing a person would need before using it.

At the end: a table with the scores, the totals, which one you would choose and why, and a list of the concrete claims I should verify before using either.

ORIGINAL TASK:
[PASTE THE PROMPT YOU USED]

ANSWER A:
[PASTE ANSWER A]

ANSWER B:
[PASTE ANSWER B]

With tool, plan and model chosen, what comes next is getting the most out of them with well-written instructions. That is module 4, "Prompts that work the first time".