Search interest here is wildly lopsided: ChatGPT dominates, Gemini is a clear second, and the rest are a long way behind. That ranking tells you what people try first, not what suits your business.
The more useful sorting is where your work already lives, because three of these five are really arguments about proximity rather than capability.
Start with the free tier of whichever assistant already sits next to your work — Gemini if you are on Google, ChatGPT if you are on neither, Copilot only if you are already paying for Microsoft 365. Add a second tool only when you can name what the first one failed at. Perplexity is the clearest case for a second subscription, because sourced research is a genuinely different job from drafting.
The picks that stand out
ChatGPT
The broadest capability in one place — writing, research, images, data — and it makes no assumption about where your files live. The one to pick when you cannot predict what you will need.
Gemini
Inside Gmail, Docs and Android, so there is no copying between windows. Cheapest entry price of any assistant here, though the paid tiers are storage plans with assistant limits attached.
Claude
Holds a long document in view better and tends to stay inside the instructions you gave it. The choice when the work is analysis-heavy rather than varied.
Perplexity
Cites what it used, which is the whole point for research you will quote to a client. Narrower than a general assistant, and the public pricing shows no monthly option.
Microsoft 365 Copilot
Sits inside Word, Excel, Outlook and Teams. The only tool here with no free plan, and it is an add-on that assumes you already pay for Microsoft 365.
Mistral
A European assistant with a free plan on web and mobile, which shortens the procurement conversation when a client asks whether data leaves the EU. It also handles French with less prompting. The trade is a smaller third-party ecosystem than the US incumbents.
The prices are not comparable, and that is the trap
ChatGPT Plus and Claude Pro are assistant subscriptions. Gemini's paid tiers are Google One storage plans with higher assistant limits attached. Copilot is an add-on that assumes an existing Microsoft 365 subscription underneath it, and its advertised annual rate is shown struck through from a higher figure, so it is a promotion rather than a baseline. Four different things wearing the same kind of price tag.
Four of the five have a real free tier
ChatGPT, Gemini, Claude and Perplexity all have permanent free plans rather than trials, which makes an honest week-long test cost nothing. Copilot is the exception: there is no free plan, so evaluating it is a paid decision from the first day and its entry price sits on top of a subscription you must already hold.
What none of them fixes
If the slow part of your work is re-reading before you send rather than producing a draft, an assistant is the wrong purchase and a proofreading tool is the right one. And if you need the answer traceable to a source you can show a client, most of these will produce a confident answer without one — that is the specific gap Perplexity exists to fill.
Switching costs almost nothing here
This is the rare category where a wrong pick is cheap to undo. There is no database to migrate, no team to retrain, no integration to rebuild — you open a different tab. That should change how long you deliberate. Pick the one nearest the work you already do, use it for a month on real tasks, and move if it irritates you. Spending three weeks comparing them costs more than the mistake you are trying to avoid, and the comparison you run in your own work is better evidence than any table.
The real cost of running two
Two assistants means two histories. The thread where you explained your client, your constraints and how you like things phrased lives in one of them; the other starts from nothing every time you open it. That context is the thing that makes an assistant feel useful after a fortnight, and it does not transfer. It is why the defensible pairing is two different jobs — drafting in one, sourced research in the other — and why two general assistants usually means paying twice to rebuild the same context.
Pricing side by side
| Tool | Verified price | Billing condition | Free | Checked |
|---|---|---|---|---|
| ChatGPT | €23.00/mo · Plus | monthly billing | Yes | 05.09.2026 |
| Gemini | €4.99/mo · Google AI Plus | per person/month; the paid tiers are Google One storage plans with Gemini limits attached, not a standalone assistant subscription | Yes | 08.08.2026 |
| Claude | $20.00/mo · Pro | $20 if billed monthly; $17/mo billed annually ($200 up front) | Yes | 06.08.2026 |
| Perplexity | $17.00/mo equivalent · Pro | USD 17/month shown for annual billing; the public pricing page publishes no monthly-billing figure | Yes | 13.09.2026 |
| Microsoft 365 Copilot | $25.20/mo · Microsoft 365 Copilot Business | USD 25.20 per user/month billed monthly; USD 18/user/month paid yearly, itself a promotional rate shown struck through from USD 21. It is an add-on: you must already hold a Microsoft 365 plan | No | 05.09.2026 |
| Mistral | $14.99/mo · Pro | monthly billing, excluding taxes | Yes | 12.08.2026 |
Frequently asked questions
Do I need to pay for any of them?
Often not at first. Four of the five have free plans that cover real small-business work. The signal to pay is a limit that stops you finishing something specific — not a plan page implying the free tier is for hobbyists.
Why is Gemini so much cheaper at entry?
Because it is not the same product. The entry tier is a 400 GB Google One storage plan with doubled Gemini limits attached. If you need the storage it is excellent value; if you do not, the headline figure is comparing a bundle against a standalone subscription.
Is it worth running two?
Only when you can name the failure. The most defensible pair is a general assistant for drafting plus Perplexity for research you have to source. Two general assistants is usually paying twice for the same job.
How long should I test one before deciding?
About a month of real work, not an afternoon of trick questions. Assistants are easy to impress in a demo and reveal themselves on the fourth similar task, when you find out whether it remembers your constraints and how much you rewrite. Free tiers make that test cost nothing for most of the options here.
Does it matter which one has the newest model?
Less than the marketing suggests. The lead changes hands regularly and the differences at the top are smaller than the difference between an assistant that sits next to your files and one that does not. Choose on proximity to your work and on what you can try for free; the model ranking will have moved again by the time a subscription matters.