Cursor, Windsurf, Codex, and Claude Code use different combinations of subscriptions, included usage, overages, and API billing. A monthly headline price is therefore not enough to identify the lowest cost for a real coding workload.
Independent comparison note: Verdent is not affiliated with, endorsed by, or sponsored by the third-party products and providers referenced on this page. Product and company names are the property of their respective owners. Details were checked against the linked official sources; current account pages remain authoritative.
As verified on August 28, 2026, their official billing structures differ:
- Cursor: Paid individual plans include a usage pool; model usage is accounted for at published API rates, and additional usage can be purchased.
- Windsurf: Its current pricing URL routes to Devin's pricing page, which lists recurring self-serve plans, included usage, and extra usage at API pricing.
- Codex: ChatGPT plans include Codex usage. Local CLI, SDK, and IDE-extension work can also use an API key at standard OpenAI API token rates; that route does not include Codex cloud features.
- Claude Code: Pro and Max include Claude Code usage, while Anthropic Console and supported cloud-provider routes use metered token billing.
Do not convert a subscription price directly into a token price. Run the same representative repository tasks and record fixed fees, overage or API spend, accepted changes, failed attempts, and human repair time. The decision metric is cost per accepted, verified task—not cost per request.
Sources: Cursor Models & Pricing, Windsurf pricing, OpenAI Codex pricing, and Anthropic Claude Code costs.
Related reading: How to control AI coding-tool costs and How credits work in AI coding tools.
