The 2026 flat → usage pivot
The coding-agent market's pricing earthquake: GitHub Copilot moved from flat per-seat to usage-based credits in mid-2026, triggering bills that jumped from $50 to ~$3,000. Cursor held flat and took share. Every row is dated and sourced.
| Vendor | Event | Detail | When | Basis | src |
|---|---|---|---|---|---|
| GitHub Copilot | Flat tiers | $10 Pro · $19 Business · $39 Enterprise | 2022–early 2026 | Per seat (unlimited) | ↗ |
| GitHub Copilot | Pro+ added | $39/mo Pro+ tier introduced | Feb 2026 | Per seat | ↗ |
| GitHub Copilot | Usage-based announced | $0.01 / premium request credit beyond included quota | Apr 27 2026 | Per credit | ↗ |
| GitHub Copilot | "Flex" pivot | Pro $15 · Pro+ $70 · new Max $100/$200 (included credits) | May 12 2026 | Seat + included usage | ↗ |
| GitHub Copilot | Usage billing live | Overage billing goes live; backlash over $50→~$3k bills (Discussion #192948) | Jun 1 2026 | Per credit | ↗ |
| Cursor | Flat pricing held | $20 / $60 flat; $2B ARR (Feb 2026); Composer mode | Feb 2026 | Per seat | ↗ |
| Windsurf | Added quotas | $15 flat but added daily/weekly request quotas | Mar 19 2026 | Seat + quota | ↗ |
| Roo Code / Cline / OpenHands | BYOK model | $20–40/mo, raw API pass-through (bring your own key) | 2026 | Pass-through | ↗ |
How coding agents moved from flat per-seat tiers to metered usage, and who it hit
The billing shift is the same movement seen from the vendor side: named events where a company moved from seat-based to usage or outcome-based pricing, with the date and the source.
Show more
These changes are hard to reverse and therefore informative. A vendor that prices per resolution is asserting its agent resolves things reliably enough to underwrite. Expect the shift to continue where the unit of work is countable, and to stall where it is not.
Why did my AI coding tool bill go up?
Because most of the major vendors moved from a flat monthly fee to usage-based billing. Under a flat fee, light users were quietly subsidising heavy ones. Metering removes that, so anyone running a lot of agent work now pays what it actually costs.
What is usage-based pricing for AI tools?
You are charged for what you consume rather than for having an account. Each model call draws down a credit balance or is billed per token. Two people on the same plan can pay very different amounts.
Are usage-based AI tools more expensive?
Not for most people. Metering usually lowers the bill for light and average use, and raises it sharply for the heaviest users. Whether you win or lose depends entirely on your own call volume.
Why are AI coding tools moving away from flat pricing?
Because agentic coding burns far more tokens than autocomplete ever did. One instruction can run for hundreds of model calls across many files. A fixed monthly fee stopped covering the cost of the heaviest users.
Can I cap what an AI coding tool costs me?
Usually yes. Most usage-based tools let you set a hard spending limit or buy credits up front. Check whether hitting the cap stops the tool or simply keeps billing you, because the two behave very differently mid-task.