Quick answer
Copilot meters premium requests or credits depending on plan, Cursor meters requests and premium model usage against an allowance with on-demand billing, and Claude Code meters tokens over a session window plus a weekly cap. The units are not interchangeable, so compare cost per completed task rather than plan price.
The three meters
| Copilot | Cursor | Claude Code | |
|---|---|---|---|
| Unit | Premium requests or credits | Requests and model usage | Tokens |
| Allowance | Monthly per seat | Per billing period | Session plus weekly |
| Overflow | Overage where enabled | On-demand billing | Extra usage or API |
| Cheapest mode | Standard model | Auto | Smaller model |
Choose by workflow
Completion-heavy work suits Copilot's model. Mixed editor and agent work suits Cursor. Long autonomous coding suits a token-based plan like Claude Code. If you use more than one, track them together: UsageAtlas covers all three, and the value formula makes the comparison honest.
Read a price change before reacting
When a vendor changes a plan, separate three things: the price, the allowance, and the meter. A price increase with a larger allowance may cost the same per task. A same-price change that moves from requests to credits can cost more for agent work while costing less for completions.
Measure the week after the change, not the announcement. Your own numbers tell you whether the effective cost moved, and a tracker that keeps history makes the comparison a glance rather than a rebuild. The value formula is the one to apply.
Questions, answered
Which is cheapest: Copilot, Cursor, or Claude Code?
Entry tiers are similar in price, but metering differs. Compare cost per completed task for your own work rather than list prices.
Do all three charge for agent usage?
Yes. Agent runs consume the allowance faster than chat or completions on all three.
Can I use two of them?
Yes, and many developers do: one for completions, one for long agent work. Track both to see which earns its fee.