The short answer
ChatGPT is the assistant most people should pay for: it covers the most ground from one prompt box, and its free tier is not a trial. Gemini is the value pick from $4.99/month. Mistral Vibe is the free plan worth staying on. Claude is what to buy if the work is code.
The best AI assistants have converged. All eight we looked at write, search, read a file and run code, and the differences that decide a purchase now sit in the price ladder, in what each vendor will tell you about usage, and in what happens to what you type.
We checked every figure against the vendors' own pricing pages and help centres on , the date stamped on all eight entities here. We did not run a hands-on trial of every plan and we do not claim to have.
Why the best AI assistants are not ranked in score order
Claude scores 8.4 and Gemini scores 8.3. Both sit above ChatGPT at 8.2, and ChatGPT still takes Best Overall.
That is not an oversight. The slot answers "what should most people put on a card", and the answer turns on breadth and on how far the free tier gets you before the card comes out at all. Claude wins the criteria because a terminal coding agent and an agentic desktop worker are included at the entry price, which is genuinely more product for the money. Most people do not have a terminal open. A score is a property of the product, not a queue position.
The same logic runs the other way. Perplexity scores 7.7, above our Best Free pick, and it is outside the four. It answers questions from the live web rather than carrying a piece of work, and its free plan is too small to work on.
Where the ladder actually starts
Every comparison of this category quotes one number per vendor and it is almost always twenty dollars. That is the least useful part of the market, because everyone has a twenty dollar tier and they are close to identical.
The real ladder, cheapest paid plan first: Gemini at $4.99, ChatGPT Go at $8, Microsoft 365 Personal at $9.99, Mistral Vibe Pro at $14.99, Claude Pro at $20 on annual billing, Perplexity Pro at $20, and SuperGrok at $30.
Two things fall out of that. Below fifteen dollars you are not buying a cheaper version of the same thing, you are buying a bundle or a smaller assistant: Gemini's entry tier is mostly storage, Go withholds the reasoning models, and Microsoft 365 Personal is an Office subscription with a small AI allowance attached. Above the twenty dollar shelf, the ladder stops being about features at all. OpenAI, Anthropic, Google and Perplexity each sell a hundred dollar tier and a two hundred dollar tier whose stated difference from the one below is how much you are allowed to use.
Almost nobody will tell you how much you can use
OpenAI, Anthropic and Google all express a paid allowance as a multiple of another plan. Two times, four times, five times, twenty times. No arithmetic turns a multiple into messages, uploads or research runs, because none of the three publishes the number being multiplied.
Perplexity is stranger. It states exact ceilings on the free plan and on both enterprise tiers, and states none on Pro and Max, which are the two plans most people buy.
Microsoft Copilot is the one that answers the question. Microsoft 365 Personal includes sixty AI credits a month, ten minutes a day of Vision and thirty minutes a day of Voice. Those numbers are small, and a small number you can plan against is worth more than a large one you cannot see. Weight that accordingly.
The free tiers, and what each one takes away
Free is not one thing here. ChatGPT gives the most, unlimited everyday text chat, with separate and unpublished caps on uploads, image creation, voice and research. Perplexity gives the least and is the most honest about it, at three Pro Searches a day and one research query a month. Microsoft's free Copilot works with no account at all.
Mistral Vibe takes the slot for what happens next. Its free plan includes ten dollars a month of API credits, which nothing else here offers, and the step up is the cheapest paid assistant in this comparison. The catch belongs in the same sentence: free tier conversations are used for training by default, and every paid Mistral tier is not.
What each one does with what you type
Treat the training default as part of the price. It differs per vendor and it is checkable.
Microsoft states it does not use prompts, responses or file contents to train foundation models, on consumer and commercial plans alike, with no setting to change. That is the strongest position in the set. Claude makes it a choice and attaches five year retention when the choice is left on. Gemini keeps chats eighteen months by default, and up to three years when a chat goes to a human reviewer, and switching Keep Activity off still leaves seventy two hours. Mistral trains on its free tier only. Perplexity trains by default and states plainly that opting out does not remove what was already used. All of that comes from the vendors' own policies, and all of it is scored under trust in how we rate.
DeepSeek is the one we do not recommend. Its own policy says data is collected, processed and stored in China, gives no retention period, and names no control for objecting to training use. Two national regulators acted against the app in its first weeks. It is capable and free, and neither of those facts is the one that decides it.
