Chatbots, assistants and models: who's who
Make sense of the names you keep hearing, so you know what you're choosing between.
There are three words you'll hear all the time:
- The company (sometimes called a lab): who makes it. For example OpenAI, Anthropic, Google, Microsoft and Meta.
- The app (the assistant or ): what you actually use. ChatGPT, Claude, Gemini, Copilot and Meta AI are all apps.
- The : the "brain" inside the app. Each company has several, with names like Claude Sonnet or GPT-6. Newer, bigger models are usually smarter but may need a paid plan.
A simple way to picture it: the company is the car maker, the app is the car, and the model is the engine. You can drive without knowing the engine, but a better engine makes a better drive.
The big apps all do similar everyday jobs well. The differences show up in harder tasks, how much you can use for free, and extras like voice or image making.
Our league table ranks the models in plain English, and shows which ones you can use for free.
Three layers
Almost every AI product you'll come across has three layers.
1. The company (lab). The organisation that builds the AI. The best known are OpenAI, Anthropic, Google, Microsoft, Meta, xAI, Mistral and DeepSeek. People in the industry call the ones that train their own models "labs".
2. The app (assistant or ). The thing you open on your phone or in your browser. ChatGPT is OpenAI's app. Claude is Anthropic's. Gemini is Google's. Copilot is Microsoft's (and it's also built into Windows, Word and Outlook). Meta AI lives inside WhatsApp, Instagram and Facebook.
3. The . The engine inside the app that actually reads your request and writes the answer. Companies release new models every few months and usually offer several at once: a fast, cheap one for everyday chat and a bigger, more capable one for harder work. Names change often, which is why haystak ranks them for you each week.
Why the model matters
Two people using "ChatGPT" might be using different models, depending on whether they pay and which option they pick. The better models are noticeably stronger at long documents, tricky reasoning, coding and careful writing. For a quick email or recipe idea, a free model is usually plenty.
Other names you'll hear
- AI tool: an app built for one job, such as making presentations, transcribing meetings or editing photos. Many tools use a model from one of the big labs inside. See our tools rankings.
- Built-in AI: features added to things you already use, like "help me write" in Gmail or Copilot in Word.
- Open model: a model the company lets anyone download and run. Popular with developers; you don't need to worry about it to get started.
So which should you use?
For everyday use, any of the big four (ChatGPT, Claude, Gemini, Copilot) is a good start, and all have free versions. The next lesson helps you pick one. If you already use Google or Microsoft products a lot, their assistant may fit neatly into what you do.
The landscape in more detail
The market is dominated by a handful of companies that train their own frontier models: the most capable systems available. In late 2026 the leaders are OpenAI, Anthropic and Google, with Meta, xAI and several Chinese labs (including DeepSeek, Alibaba and Moonshot) close behind on many measures. Our league table tracks where each one stands every week.
Model families and tiers
Most labs release a family of models at once, in tiers. Anthropic, for example, has Haiku (fast and light), Sonnet (balanced) and Opus (most capable). Google and OpenAI use their own names for similar tiers. The pattern is the same everywhere:
- Small and fast tiers are cheap to run, so they're often what free plans use.
- Large tiers are better at long, complex or careful work, and usually need a paid plan (typically around £18 to £20 a month for individuals).
- Reasoning or "thinking" modes take extra time to work through a problem step by step before answering, which helps with maths, planning and tricky analysis.
Version numbers rise quickly, and older versions are retired. That's why haystak flags models that are retiring soon.
How apps and models relate
An app can switch the it uses behind the scenes, offer you a menu, or pick automatically depending on your question. Paid plans tend to unlock the stronger models, higher daily limits and extras like longer documents, voice conversations, image generation or "agent" features that carry out tasks for you.
Some apps don't make their own models at all. Microsoft Copilot has historically used OpenAI's models alongside Microsoft's own. Thousands of specialist tools quietly use a model from one of the big labs, which is why a note-taking app or a writing tool may "feel" similar to ChatGPT or Claude.
How models are compared
There's no single perfect measure. Common approaches include:
- Benchmarks: standard tests, such as exam-style questions, coding challenges or long-document tasks.
- Blind head-to-head votes: people compare two anonymous answers and pick the better one, building up a ranking.
- Price and access: whether you can use it free, and what it costs.
haystak combines independent benchmark results and blind-vote results into one score out of 10, explained on our methodology page. Rankings at the top can be very close, so treat a difference of a few tenths as "about the same".
What actually matters for you
For most non-technical uses, the gap between the top few assistants is smaller than the gap between a vague request and a clear one. Choosing a sensible app and learning to ask well (lesson 7) will improve your results far more than chasing the newest model every month.
Try this prompt
Explain the difference between ChatGPT, Claude, Gemini and Microsoft Copilot in a short table: who makes each, what it's best known for, and whether there's a free version.
Replace the [brackets] with your own details
On haystak
Quick check · 1 of 3
In "Claude Sonnet", what is Claude?
Next step
Pick your first free assistant and set it up safely
Choose one free assistant and set it up in 10 minutes, with the privacy settings that matter.