L'article ci-dessous est publié en anglais. Chaque chiffre qu'il contient est le chiffre publié par l'éditeur lui-même, avec la date à laquelle nous l'avons lu.
Three assistants, one question: what does it cost to stop being on the free plan?
Why this axis and not quality
We do not run these models against each other on the same prompts and score the output, so we are not going to rank them on it. The axis we can rank, and source, is the one every buyer reaches second: after the free tier stops being enough, what is the smallest amount the vendor will take.
Why these three numbers are not three prices of the same thing
They are all dollars per month for the cheapest paid plan, which is close enough to rank. They are not close enough to treat as interchangeable, and the differences are the useful part of this page.
Google's figure buys a plan it calls Google AI Plus rather than a Gemini plan. OpenAI's figure is marked, by OpenAI, as the price displayed in the United States, with pricing localized in some markets. Anthropic's figure is the monthly one, and its card leads with a lower rate for paying a year up front.
The three vendors also treat the yearly question three different ways, which is worth knowing before you convert anything. Anthropic prints an annual rate beside the monthly one. OpenAI states in its own help center that it does not support annual billing, or paying for several months in advance, on any of its consumer tiers. Google's plans page prints monthly figures and does not raise the question at all, which is a fact about that page and not an answer.
The axes we scored and dropped
Free tier was the obvious one, and it does not separate these three. All three publish a free plan, and only OpenAI states in words what it grants: Anthropic's card and Google's card both print a price and no allowance. An axis that splits a field one against two, with nothing published to break the tie, cannot order three products.
Context window and message caps were dropped for different reasons. On the pages we read, Anthropic publishes no message count at all and Google publishes none either; OpenAI publishes counts only for its two highest Pro models (50 messages a week on Pro $100 and 200 on Pro $200 for GPT-6 Pro in chat), which is not a figure a buyer of an entry tier can use. Context windows are not comparable either: Anthropic's pricing grid prints "Up to 1M varies by model" per tier, and its help article says that on paid plans the newest models support up to a 1M token context window while others support 500K or 200K, whereas the other two do not print a figure on their cards. Ranking on figures only one vendor publishes, or on a number only the top tier of one vendor carries, would have required normalizing different measures into one, which is a judgment we chose not to invent. The ChatGPT review and the Claude review each read their vendor's usage documents in full, and the Gemini review reads Google's usage multiples and the privacy setting the plan names sit beside.
How to read the ranking
Position one is not the best assistant here. It is the cheapest way to start paying, on the day we read the pages, in the country the page was written for. For most readers the free tier decides this and the price never comes up. The profile behind each entry carries the figures and the documents they came from, and those are what a decision should rest on.