Point opencode, Cline or any OpenAI-compatible tool at one key. Your plan gives you a dollar allowance that resets every 5 hours, and every model draws from it at its normal list price.
All 31 models in the catalogue are included. No per-model plan, no seat juggling, no upgrade maze.
Requests per 5 hours assumes 7,500 uncached input tokens, 42,500 cached input tokens and 1,000 output tokens per agent request, at the list prices of deepseek-v4.1-flash-1017. If a model has no cached price, its input price is used. Rounded to the nearest 10 above 100.
Prefer no subscription? Pay as you go stays available at the same list prices.
| Plan | Price | 5-hour allowance | Weekly allowance | Plan keys | Rate limit | Seats | Requests / 5 hours |
|---|---|---|---|---|---|---|---|
| Lite For one agent and evening projects. |
$10 / 30 days | $0.50 | $3.00 | 1 | 3,000 / min | Unlimited | 590 |
| Pro For daily work with a coding agent. |
$25 / 30 days | $1.50 | $8.00 | 2 | 3,000 / min | Unlimited | 1,780 |
| Max For several agents running in parallel. |
$60 / 30 days | $5.00 | $18.00 | 5 | 3,000 / min | 20 of 20 seats left | 5,930 |
Each request is priced with the same list-price cost function as pay as you go. The plan pays that amount instead of your prepaid wallet.
Your allowance resets every 5 hours, with a weekly cap anchored to the subscription start. Nothing rolls over.
Requests bill from prepaid balance if you have one. Otherwise they wait for the reset. Prepaid balance is never touched while the plan has allowance.
Works with opencode, Cline, SillyTavern, LiteLLM, Open WebUI. Any OpenAI-compatible client works.
All 31 models in the catalogue. There is no per-model plan and no separate allowance for a new model.
New models are in the plan immediately, at their normal list price. You do not need a new plan or a new key.
Yes. A change takes effect at the next renewal. A cancellation runs to the end of the paid period, then the account returns to pay as you go. Prepaid balance is never touched.
Max is capped at 20 active seats so parallel-agent capacity stays predictable. When it is full, the Subscribe button reports the remaining count.