Both put a model inside the place you already write code. The difference that decides it is not the model, because both let you point at several. It is how much of your existing setup you are willing to hand over, and what each one tells you before you pay.
What the table separates
The two rows marked with a source carry published figures and name the document each came from, so you can check both in about a minute. The rows marked Editorial call are our judgment, and each one says what the judgment is a read of. We have not run the same task through both tools and scored the output, so nothing here is a quality comparison.
The free tiers are not the same kind of promise
Both tools have one, and only one of them is countable. GitHub publishes a number for Copilot Free in its billing documents: 2000 completions a month, plus a monthly allowance of GitHub AI Credits, with chat billed in credits. Anysphere's Hobby card says Limited, applied to Agent requests, with no digit beside it on the pricing page or in the help doc.
That is a difference in what is published, and it is worth keeping apart from a difference in what is granted. Cursor's free tier may well be the more generous of the two. We cannot tell you, and neither can anyone who has only read the page.
What is not in this comparison
Neither tool's overage behavior is in the table, and it is the thing most likely to decide the second month rather than the first. Cursor's docs describe included usage as a monthly allowance across two pools, one for its own models and one for third-party models, with both columns marked Included but no amount printed for either; when included usage runs out, on-demand usage can be enabled and is billed at the same model API rates with no markup. GitHub prices its credits individually at a named rate, 1 AI credit = $0.01 USD, with a dollar-denominated budget an account can set for additional usage. Both are real answers to the same question and they are denominated differently, so we have left them on the profiles rather than putting two numbers side by side that do not convert.
The entry prices in the table are the individual plans. GitHub also prints per seat figures for organizations, now in the table above, and its documentation says the individual price is free for some users, naming verified students on a distinct Copilot Student plan and verified teachers and maintainers of popular open source projects on Copilot Pro. If you are one of those, the row above is not what you would pay.
One more split decides the purchase for teams: the cloud agent on GitHub.com is gated at Copilot Pro, so the free tier is the only individual plan without it, and third-party coding agents begin at Pro. Cursor, being the editor itself, offers its agent on the Hobby plan for free, described as limited. The agent story is therefore the reverse of the entry-price story: Copilot is cheaper and Cursor's agents reach further on the free tier.
Both profiles carry the same figures with the same dates: Cursor and GitHub Copilot, the review of GitHub Copilot covers the organization privacy answer, and AI Coding is the category hub. The Cursor review reads the editor-purchase side: the two pools, Privacy Mode and what leaving costs.