Subscription ROI methodology
How we estimate what a plan's usage would cost at API prices. The catalog has 52 plans, 30 of them direct.
Formula
ROI = (API value − plan price) / plan price × 100%. Savings $ = API value − price. Break-even = price / API value — the share of limits you must use for the plan to beat the API.
Direct subscriptions (usage-based)
For plans from model providers (ChatGPT, Claude, Gemini, Grok, GLM Coding Plan) we take published limits — messages per window, weekly quotas, Codex or Claude Code credits, Deep Research runs — multiply messages by a typical prompt and response size and by the model's official per-million-token price, including prompt-cache discounts. When limits are not published, we use a conservative lower bound from the official help pages.
Proxy subscriptions (credit-based)
For IDEs and aggregators (Cursor, Windsurf, Copilot, Trae, Kilo) the value equals the included usage credits spent at the underlying models' API prices. Bonuses and unlimited modes on small models are not counted.
Assumptions
- A typical request is 2–8K input and 0.5–1.5K output tokens; agent runs use more context but a higher cache share.
- A working month is 22 active days, not round-the-clock limit use.
- API prices are the providers' official list prices on the check date, without volume discounts.
- Temporary promotions and bonus limits are excluded.
FAQ
What is a plan's API equivalent?
It is what you would pay at official API prices for the usage a plan includes under active but realistic work.
Why can real value differ?
Many plans have dynamic limits that depend on context length, model choice, and load. If you use only part of the limits, the value shrinks proportionally.
Are promotions included?
No. Temporary promotions, bonus limits, and annual-billing discounts are excluded unless stated otherwise.