Case study: Ezoic
About $450,000 in July token spend. Projected: under $25,000 a month.
Ezoic reported spending about $450,000 on model tokens in July. With Pumpkin, Ezoic projects its token spend will fall below $25,000 a month through subscription quota, Auto model choice, context summaries, and no Pumpkin own-key token fee.
The numbers
Both figures are token spend: what model providers bill per token for API use. The first is Ezoic's own approximate figure. The second is Ezoic's forecast, not a measured result.
Subscription quota replaces paid API use
A Codex or Claude Code subscription includes a quota of model use for a fixed monthly price. An API key bills every token. When agent work runs through API keys, every request adds to the token bill.
Pumpkin connects to the Codex or Claude Code app you are signed in to. By default, eligible requests use subscription quota first. When quota runs out, Pumpkin can use a paid API route if you configured one. Work that would be billed per token runs on quota you already pay for.
The forecast is Ezoic's own. Your numbers depend on your plans, your models, and how much work you run.
Auto matches capability, then favors lower cost
Auto mode first matches the task to a model that can do the work. Among suitable models, it favors lower cost. It does not promise one universally cheapest or best model.
Context keeps task facts and recent work
Context management turns older exchanges into a shorter summary while task facts and recent work stay in context. This reduces repeated input; saved history remains available when details need another look.
No Pumpkin per-token fee on your own keys
Pumpkin adds no per-token fee for your own-key input, output, or cache tokens. Provider prices, including provider cache prices, still apply.
Cursor charges Teams and Enterprise $0.25 per million input, output, and cached tokens, including own-key requests. The fee draws from included usage; further charges require on-demand usage. Checked October 13, 2026.
What the figures cover
Token spend only. Codex and Claude Code subscriptions are a separate cost paid to the providers, and Pumpkin seats are a separate $20 per person per month. Provider plan terms and limits still apply. A subscription's quota is not a cash balance, and a configured API route costs money when it takes a request. Pumpkin's usage reports show which route served each request, so you can see subscription-backed work apart from API work.
Do the same with your plans
- Install and sign in to the Codex or Claude Code app on each machine.
- In Pumpkin, open Settings, then Models, and add the subscription models.
- Check Provider order: subscriptions come first by default. Add an API key only if you want a paid route when quota runs out.
How subscriptions work in Pumpkin. How Auto chooses models. How chat context works. Pricing lists the seat price and who bills what.