How to evaluate an AI coding agent's pricing model
Seat-based, usage-based, and hybrid pricing all show up in this category — here's what each means for a team's actual monthly bill as usage scales.
Crail Editorial · Published 2026-07-18 · Last verified 2026-07-27
AI coding agents price themselves in at least three different shapes, and comparing a “$20/month” tool to a “$0.03/1K tokens” tool on the sticker price alone is close to meaningless. Here’s what to actually look at.
The three pricing shapes
Per-seat, flat. A fixed monthly fee per developer, usually with soft usage limits (e.g. GitHub Copilot’s Pro tier at $10/month, or Claude Code’s Pro plan at $20/month with a token-usage ceiling). Predictable, easy to budget, but you’re paying the same whether a seat is used heavily or barely touched.
Usage-based / credit-based. You pay for what you consume — tokens, agent runs, or “compute units” — sometimes with a bundled monthly credit allowance and pay-as-you-go beyond it (Cursor’s Pro tier bundles a $20 credit pool; Devin bills on Agent Compute Units on top of a subscription). This rewards light usage and can get expensive fast for teams running agents continuously or on large codebases.
Hybrid. A base subscription plus metered overage — the most common shape as of 2026, since it gives vendors predictable base revenue while letting heavy users pay proportionally more.
What actually drives the real bill
- The overage rate, not the sticker price. A $20/month plan with a punishing overage rate can cost more in practice than a $39/month plan with generous included usage, once a team’s actual agent-run volume is accounted for.
- Whether usage is pooled or per-seat. Team plans that pool usage across seats are more forgiving of uneven usage than ones that hard-cap each seat individually.
- Whether you can see the meter. Tools with transparent, real-time usage dashboards let you catch a cost spike before the invoice; tools without one don’t.
- Self-serve vs. sales-call pricing. Enterprise-only, quote-based pricing (common for tools like Sourcegraph Cody’s current enterprise-only offering) means your real price is a negotiation, not a published number — budget accordingly and expect it to vary by deal size.
Where to check this on Crail
Every vendor page under AI Coding Agents publishes the full tier breakdown — including overageRate where a vendor discloses it — plus whether the tool requires a sales call to buy at all, sourced directly from the vendor’s own pricing page with a last-verified date.
FAQ
Is usage-based pricing more expensive than per-seat?
It depends entirely on usage patterns — usage-based can be cheaper for light users and much more expensive for heavy ones. Check the overage rate, not just the headline price.
Why do some tools require a sales call just to see pricing?
Usually because pricing is negotiated per-deal at enterprise scale (common for legacy enterprise software); Crail flags this as requiresSalesCallToBuy on every vendor record since it also affects how usable a tool is for an autonomous agent doing the buying.