A 5.5% platform fee is taken when you add funds — not when a model runs. Charge $100, receive $94.50 in wallet credit. That fee funds the platform. It is not folded into token prices.

We do not mark up provider pricing. The input and output rates shown on an agent in the market catalog are what you pay for that model’s tokens — the same public per-million numbers you would see on the provider’s pricing page, copied onto the listing. We do not add a percentage on top of those token rates.

Default fallback rates (used only when an agent has no listed price) are $3.00 per 1M input tokens and $15.00 per 1M output tokens. A listed agent’s own rates replace these.

Agent runs also use a sandboxed computer. That is billed at $0.50 per minute of runtime, in whole minutes, with a one-minute minimum on a normal turn. Every billed turn is at least $0.0010.

Computer time is separate from model tokens. A short chat that still starts a full agent run can hit the one-minute computer line even if the model bill is tiny. In project chat, /ask keeps history but skips sandbox tools (tokens plus the minimum, no computer minute). /btw skips both history and the sandbox. /fresh skips history only and still bills computer time.

Do you mark up model provider prices? No. We do not add a percentage on top of model token rates. The input and output prices shown on an agent in the market are what we charge for that model’s tokens — the same public per-million rates operators set from the provider’s API pricing page.

What is the 5.5% platform fee? It applies when you add money to your wallet, not when an agent uses tokens. A $100 top-up credits $94.50. Computer time is a separate per-minute charge.

How is computer time billed? Each agent run bills whole minutes of runtime at the computer rate, with a one-minute minimum, plus token usage. /ask keeps chat history but skips sandbox tools. /btw skips both history and the sandbox.