Claude Code vs Cursor vs Codex vs Gemini: the revenue table

Claude Code leads on revenue evidence — 49 indexed cases, 10 third-party verified — with Cursor right behind at 44 and 10. But this table ranks which tool's builders publish numbers, not which tool earns more. Three files are too thin to rank: Windsurf has 2 cases, Copilot 5, Codex 10. Read those three as anecdotes.

Contents

• ·Which AI coding tool has the most revenue evidence?
• ·What does each headline tool's file say?
• ·What about the prompt-to-app and thin files?
• ·How is a case graded?
• ·Which one would we actually pick?
• ·Frequently asked questions

Which AI coding tool has the most revenue evidence?

Claude Code, on both counts: 49 indexed cases and 10 third-party-verified revenue figures, against Cursor's 44 and 10. Codex is proportionally better verified — 3 of 10 — and far too small to rank. Bands below come only from cases publishing a clean monthly figure.

Grades, first use: ✅ third-party verified — a dashboard, store payout or tracking tool shown on camera. 🗣 founder-reported — the operator said it. 📎 creator-relayed — a channel passed on someone else's number. 🔮 unproven — no figure at all, just a target.

What does each headline tool's file say?

Claude Code and Cursor are statistically twins; Codex and Gemini are not comparable to them or to each other. The two big files produce nearly identical medians from identical verified counts. The two smaller ones differ in kind, not only in size.

Claude Code (49 cases, 10 ✅). The widest spread we hold: $100/mo for Fluently up to $500K/mo for the CoinSnap-style identifier portfolio, both verified. It also carries the most 🔮 entries — ten, mostly service offers and unreleased apps. Also: apps built with Claude Code.

Cursor (44 cases, 10 ✅). The tightest middle in the index: 18 of 26 monthly figures land between $1K and $20K/mo. Cases skew toward consumer apps and single-workflow tools — Launch Fast at ~$21.8K/mo in 90 days is the shape that repeats. See apps built with Cursor.

Codex (10 cases, 3 ✅). The strangest column here. Not one Codex case publishes a monthly figure under $20K/mo, which almost certainly means small Codex projects exist and nobody filmed them. Its 3-of-10 verified rate beats Claude Code's; ten rows cannot carry that conclusion. See apps built with Codex.

Gemini (26 cases, 3 ✅). Gemini usually appears as the *model* inside a product, not the agent that wrote it — nano-banana.ai at roughly $115K/mo net 📎 is generation, not engineering. Only 3 of 26 are verified, the weakest ratio among the big files. See apps built with Gemini.

What about the prompt-to-app and thin files?

Lovable, Bolt and Replit index 47 cases between them but only 16 clean monthly figures, and Windsurf and Copilot together hold 7 rows — two of which are the same vendor's balance sheet. These five columns measure published attention, not earning power.

Lovable (19 cases, 4 ✅). Its largest reported case, Grower, is $180K/mo 🗣 — but the only verified app built with Lovable is Fluently at ~$100 MRR. Three of the four ✅ records are category scale references, including Base44's $80M exit. Background: apps built with Lovable.

Bolt (15 cases, 2 ✅). A small file with a high floor: nothing below $10K/mo, topped by Sprout at $250K/mo ✅ and the influencer-partnered venture studio at ~$67K/mo ✅. Still only two verified records out of fifteen. Apps built with Bolt.

Replit (13 cases, 3 ✅). Nothing clears $100K/mo. The verified app someone actually shipped on Replit is the Peptide Tracker at $11K MRR; the other two ✅ records are Replit and Base44 as companies. Apps built with Replit.

Windsurf (2 cases, 1 ✅). Two rows, one of them Windsurf itself at a $1.3B valuation — a vendor's balance sheet says nothing about its users. See Windsurf's revenue evidence.

Copilot (5 cases, 1 ✅). Read all five and only one was actually built on Copilot. Probably the most-used tool on this page and the least narrated — its users ship at work and never film it.

How is a case graded?

Every case starts from a video breakdown with a named business and a revenue figure, then gets one of four grades. We grade the source of a number, never its size — a $100/mo ✅ record is worth more to a decision than a $1M/mo 🗣 one.

1. 1.✅ verified needs corroboration beyond the founder's word: a dashboard on camera, a store payout, a Sensor Tower placement.
2. 2.🗣 founder-reported is the operator's own unaudited claim.
3. 3.📎 creator-relayed repeats someone else's figure.
4. 4.🔮 unproven covers targets, price lists and projections with no cash behind them.

Two exclusions shape the bands. Benchmark numbers — what somebody *else's* app earns — never enter one, so the $1.3M/mo Floe clone and $200K/mo TinyURL reference are absent. Peaks, lifetime totals and valuations are out too: Cursor's own $500M/yr is a scale reference, not a case. Method and index at /blog/ai-coding-tools.

Which one would we actually pick?

Pick on what you are building, because this data does not separate Claude Code from Cursor at all — same verified count, medians a thousand dollars apart. Across all nine files, the tool looks like the least load-bearing decision anyone made.

Our position: for a consumer app with one obvious action, the Cursor column has the most examples of that shape working. For a service or an SEO asset, Claude Code has more, including Subscribr at $30K/mo ✅. Non-technical? The prompt-to-app tier reaches a paying customer faster, but its verified evidence is weakest here — Lovable's best-proven app earns $100 a month. Do not argue Codex, Windsurf or Copilot from this page.

More in [AI coding tools](/blog/ai-coding-tools): every pairwise comparison above, plus the full revenue ranking, what Claude Code is and what Cursor is.

Frequently asked questions

Which AI coding tool makes the most money for its users?

None of them, measurably. Claude Code and Cursor both hold 10 third-party-verified revenue figures, and their medians sit at $15K/mo and $14K/mo — a gap smaller than the noise in any single case. What differs is what people build, and how many published a number.

Why is the Codex list so small?

Codex arrived late to the creator economy that produces these breakdowns. Ten indexed cases is an anecdote, not a sample. The oddest detail: none of the seven Codex cases with a monthly figure reports under $20K/mo, which reflects who films videos, not who ships small profitable things.

Does a verified grade mean the business is good?

No. It means the number was corroborated beyond the founder's word — a dashboard on camera, a store payout, a tracking-tool placement. A verified $100/mo extension and a verified $500K/mo portfolio carry the same grade. It says how much to trust a figure, not whether to copy it.

Should I switch tools based on this table?

Almost certainly not. In every column, the cases that earn share a narrow paid problem and a reachable buyer, not an editor. Switching costs a week of muscle memory and buys nothing the evidence can detect. The one honest signal is which community publishes numbers, not which product is better.