AI coding assistant pricing compared

Every one of these tools quotes its price differently, and one of them quotes a different price depending on who is looking. Here is the comparison, read from each vendor's own pricing page on 17 July 2026.

The individual plans

Cursor shows different prices to different people

Fetch Cursor's pricing page repeatedly and you get two different headline prices for the same plan. Across eight fetches we saw four renders showing $20 / $40 and four showing $16 / $32. The page carries an internal billing-variant flag that flips between them. This is server-side A/B bucketing, not a mistake.

The mechanism is the Monthly/Yearly toggle, and the experiment changes which position it starts in. In the $16 render, the yearly radio button is pre-selected; in the $20 render, it is not. So:

• $20 / mo. — billed monthly
• $16 / mo. — billed yearly

Both are true, and both are disclosed: the toggle is visible and you can switch it. This is a conversion experiment, not a dark pattern.

It does mean two things, though. If you and a colleague open Cursor's pricing page right now, you may see different numbers and neither of you is misreading. And any article quoting a single Cursor price is quoting one bucket — including screenshots, and including this one. Check your own cart.

The page's embedded structured data is the stable reference. In both renders it lists Pro at $20, Pro+ at $60, Ultra at $200 and Teams at $40 — the monthly figures.

Two of the five will not sell you an annual plan

Cursor, Claude Code, Gemini Code Assist and Codex Business all discount a commitment. Two do not:

• GitHub Copilot publishes monthly pricing only. The words "annual" and "yearly" do not appear on its plans page at any tier.
• OpenAI Codex's individual plans — Go, Plus, Pro — are monthly. The only annual pricing on its page is Business: $20/user/month billed annually, $25 billed monthly.

This matters when you compare. Against Cursor's yearly $16, Copilot's $10 is 62% of the price. Against Cursor's monthly $20, it is half. Same two products, different conclusion, depending on which Cursor number you happened to be shown.

Copilot's page is also the clearest of the five, and the only one that quantifies its free tier: 2,000 completions per month.

Copilot's credit mechanic is generous on paper

Copilot states plainly that 1 AI credit = $0.01, then includes more credit value than the plan costs:

Every tier includes more credit than you paid. The catch is on the page: "Flex allotments may change over time." So the flex portion — half of Max's value — is explicitly not a guarantee. Read it as a promotional allowance, not a contract.

Google publishes an hourly rate. To eight decimal places.

Gemini Code Assist is the outlier:

Subscriptions are billed monthly, in Google's own wording, with either a monthly or yearly commitment. The page has a monthly toggle — but the monthly per-seat figure did not render for us, so we are not going to multiply it out. The answer depends on the hours-per-month basis Google uses, and guessing at that would be inventing a price.

That a per-seat developer tool is quoted to eight decimal places of an hourly rate tells you which department wrote the page.

Google confirms a free version for individuals exists. Its quotas sit on a JavaScript-only page that returned no readable text, so we cannot report them. The paid editions publish real limits: 6,000 code generation requests/day, and agent mode capped at 1,500/day on Standard and 2,000/day on Enterprise.

Claude Code has no free tier

Worth stating plainly because the others do. Anthropic's free plan is chat only — Claude Code appears first at Pro, $20/month ($17 billed yearly). Max is $100 (5×) or $200 (20×).

Team seats are $20/seat yearly or $25 monthly; Premium seats $100/$125.

What we would actually do

• Compare monthly against monthly. Three of these quote a yearly-equivalent price somewhere on the page. The headline numbers are not on the same axis.
• Cheapest paid entry is Codex Go at $8, then Copilot at $10. But Go is scoped to "lightweight coding tasks" in OpenAI's own words — sort by what a tier includes, not by its price.
• Copilot is the only one that tells you what its free tier is. 2,000 completions a month. Everyone else says "limited".
• Check Cursor's price in your own account, not in an article.
• Nobody quantifies "usage limits." Every vendor here says limits apply; only Copilot attaches numbers to its free tier. Budget for heavy use costing more than the sticker.