Stop getting surprised by your agent bill.
Plans, usage limits and per-token prices for Claude Code, Cursor, Codex, Copilot, Windsurf, Devin and Gemini CLI — every number sourced from the official pricing page, with a last-verified date.
Estimate my monthly cost Get price-change alerts →Agent cost calculator
Estimate what your agent usage would cost at API list prices, and compare it with flat-rate plans. Plans don't publish token allowances, so treat plan rows as "break-even" points, not guarantees.
All models, cheapest first
Flat-rate plans vs your API estimate (selected model)
Coding agents at a glance
| Tool | Entry paid plan | Top individual plan | How limits work |
|---|---|---|---|
| Claude Code | $20/mo | $200/mo | Usage resets on a rolling 5-hour session window; paid plans add weekly limits. Claude Code shares one usage pool with Claude chat. Pro = at least 5x F… |
| Cursor | $8/mo | $200/mo | Every tier has a usage pool shared across Cursor and Grok that refreshes weekly. Plus = Grok models only. Premium and up: third-party models (Claude, … |
| OpenAI Codex | $20/mo | $500/mo | Plus: rolling 5-hour window plus possible weekly limits. OpenAI's estimate for Plus is 15-160 local GPT-6.1 Sol messages, 5-45 GPT-6 Astra, or 350-3,0… |
| GitHub Copilot | $10/mo | $100/mo | Agent mode, chat, code review, cloud agent and CLI consume GitHub AI Credits (1 credit = $0.01); credits vary by model. Allowance resets monthly. Comp… |
| Windsurf (now Devin Desktop) | $20/mo | $200/mo | No credits: quota refreshes daily and weekly. Pro/Teams published estimates: 7-27 msgs/day on Premium Plus models (e.g. Opus 4.6, GPT-5.4), 8-101 on P… |
| Devin | $20/mo | $200/mo | Self-serve plans include a quota that refreshes daily and weekly; overage is billed as dollar-denominated on-demand credits at API pricing. ACUs remai… |
| Gemini CLI | — | — | Gemini CLI + Code Assist agent mode share a quota: 1,500 requests/user/day (Code Assist Standard), 2,000 (Enterprise). One prompt can trigger several … |
FAQ
Why do I keep hitting usage limits in Claude Code?
Claude plans reset on a rolling 5-hour window with weekly caps on paid plans, and Claude Code shares one pool with Claude chat. Long contexts and Opus-class models burn the pool fastest.
Is a subscription or the API cheaper for coding agents?
It depends on volume. Use the calculator: if your API estimate for the model you use is above a plan's price and that plan's quota covers your work, the plan wins.
How current are these prices?
Every entry carries a source URL and a last-verified date; we re-check official pricing pages weekly.