The 2026 tool landscape

Four companies, four assistants

The assistant market in 2026 is dominated by four products, each from a big technology company. This whole course is built around them, so it's worth fixing the names.

• ChatGPT, made by OpenAI. The one most people met first. A general-purpose assistant, strong all-rounder, available on the web, mobile and increasingly baked into other software.
• Copilot, made by Microsoft. The assistant woven through Windows and Microsoft 365: Word, Excel, Outlook, Teams. Copilot originally ran OpenAI's models, and since mid-2026 Copilot Chat can also use Anthropic's Claude as a selectable model, often auto-selecting across providers depending on the task. The durable point is that Copilot is Microsoft's wrapper: its value is what it can see (your work files and email, when your organisation allows it), not whose model happens to answer.
• Claude, made by Anthropic. Known for careful, considered writing and for handling long documents well. Popular for drafting, analysis and anything where tone and nuance matter.
• Gemini, made by Google. Google's assistant, built into Search, Gmail, Docs and Android. Convenient if your working life already lives in Google's tools.

You don't have to pick one. A sensible way to think about them: they're four doors into the same room. Most can chat, draft, summarise, analyse a file and search the web. The differences that matter in practice are usually where they live (Copilot inside Microsoft 365, Gemini inside Google Workspace) and small differences in style and strengths, not some vast gap in raw capability.

Model families, and the picker you'll meet

Open any of these tools and, sooner or later, you'll find a dropdown of model names: things like a numbered version, or names with "mini", "pro", "flash", "opus" or "thinking" in them. This is the bit that overwhelms beginners. Don't let it. Almost every picker, whatever the labels, is really offering you a choice along one axis.

Fast versus reasoning

The single distinction to understand is fast models versus reasoning models.

• A fast model answers quickly. It reads your prompt and starts writing almost immediately. Ideal for everyday work: drafting an email, summarising notes, rephrasing something, quick questions. This is the right default for most tasks.
• A reasoning model deliberately takes longer. Before answering, it works through the problem in steps, sometimes visibly "thinking" for several seconds or more. It's better at hard problems: multi-step logic, careful analysis, thorny maths, code, planning. You pay for that quality in time, and often in tighter usage limits.

Whatever the marketing names, when you see a picker, ask yourself one question: is this a quick job or a hard one? Quick job: leave it on the fast default. Hard, multi-step problem where being right matters: reach for the reasoning option. That's most of what the picker is for.

One extra wrinkle you'll now meet: pickers increasingly carry tier names too, a provider's own brand labels for its models, and sometimes effort levels or usage-credit "frontier" options alongside them. Don't try to memorise these; the names churn, differ from provider to provider, and tell you little on their own. The question that still does the real work is the same one: quick job or hard one? That decides which to pick, whatever the label on the button says.

What free tiers include now

Here's the honest 2026 picture, kept deliberately general because the specifics shift monthly.

Every one of these tools has a free tier, and free is no longer a crippled demo. On a free account you can typically chat with a capable model, upload files and images, use web search, and get real work done. For a beginner, free is the correct place to start, full stop.

What free tiers limit rather than remove:

• How much of the best models you get. Free usually gives generous access to a fast model and a smaller ration of the newest or reasoning models; you might get a handful of the heavyweight answers per day, then drop back to the standard one.
• Volume and priority. Paid users get higher message limits and priority when the service is busy.
• The advanced extras. The newest features, deeper research modes, richer file handling, connectors to your other apps, often land on paid tiers first.

Something newer to be aware of: some free tiers have begun showing ads, which carries privacy considerations of its own; the privacy lesson (06-privacy-and-safety) explains what that means for you.

A paid tier (typically a monthly personal subscription) raises those limits, brings the newest models sooner, and adds the advanced features. There are also business and enterprise tiers, which matter enormously for work because they come with data-handling guarantees a personal account doesn't; we cover that in the privacy lesson.

Our honest advice, repeated because it's right: start free, and only pay when you can name the specific limit that's blocking you. Paying "to be safe" as a beginner just spends money on capacity you're not yet using.

It shifts every month, so check the source

Be warned plainly: prices, model names and exactly-what's-included change constantly. A new model lands, tiers get reshuffled, a feature moves from paid to free or the reverse. Any specific number in a lesson (or in a friend's advice) is a snapshot, not a fixed fact. The durable skill isn't memorising today's line-up; it's knowing to check the official pricing page before you decide.

Here are the official pages to bookmark:

• ChatGPT: openai.com/chatgpt/pricing
• Microsoft Copilot: microsoft.com/microsoft-copilot
• Claude: claude.com/pricing
• Gemini: gemini.google/subscriptions

A worked example: choosing where to run one task

Say you're at Fernway and need to turn a page of rough notes into a tidy summary. Any of the four could do it. Your reasoning might go: the notes are just text in front of me, it's a quick job not a hard one, so a fast model on a free tier is plenty, and I'll use whichever tool I already have open.

Summarise these meeting notes into six bullet points a busy manager could read in thirty seconds, then list every action and who owns it. Notes: [paste].

Now contrast a harder job, untangling a tricky decision with trade-offs, where the reasoning model earns its extra seconds:

Use your reasoning/thinking mode. We're choosing between renewing our office lease and moving to one of two new sites. Walk through the trade-offs step by step, covering cost, disruption, and staff commute, and tell me what extra information I'd need before deciding.

And when currency matters, put the tool in the right mode explicitly:

Search the web for the current pricing of ChatGPT's paid personal plan and give me the link to OpenAI's official pricing page so I can confirm it myself.

Try it now

Common mistakes

• Thinking there's one "best" tool. There rarely is. The right choice usually comes down to where your work already lives and small style preferences, not a capability league table.
• Being paralysed by the model picker. Beginners lose real time agonising over version names. Collapse it to one question: quick job or hard job? Default fast; reach for reasoning only when it's hard.
• Trusting a price or feature you read months ago. These change monthly. Over-trusting stale specifics, including confidently quoting an out-of-date price to a colleague, is an easy way to be wrong. Check the official page.
• Paying before you need to. Free tiers are capable. Upgrade when a specific limit blocks you, not pre-emptively.

Keeping current

This is the fastest-moving lesson in the phase. Treat every specific here as a 14 July 2026 snapshot and confirm current tiers, models and prices on the official pages linked above before making a decision or repeating a figure to someone else.

A few specifics that are true as of this snapshot but will move:

• Copilot and Claude. Microsoft added Anthropic's Claude (Sonnet, with Opus for heavier reasoning) as a selectable model in Copilot Chat during early 2026, alongside the existing OpenAI models. Availability and defaults vary by region and organisation: as of mid-2026 UK and EU tenants had it off by default, so what you see depends on what your organisation has enabled.
• Tier names in pickers. Providers keep rebranding their model line-ups. OpenAI's Sol, Terra and Luna tier names were introduced in July 2026, for example. Names like these are the churning label, not a durable fact; the fast-versus-reasoning question outlives them.
• Ads on free tiers. OpenAI began trialling ads in ChatGPT's free (and lower-cost "Go") tiers, in the US from February 2026 and expanding to the UK from around May 2026. Paid, business and enterprise tiers were not part of the trial. See the privacy lesson for the personalisation angle.