Plans, model allowance, and cloud runtime
Choose a monthly plan around complete research tasks. Every plan includes model allowance and sandbox runtime, with agent model calls and cloud-computer operation tracked separately.
Monthly plans
All three plans use the same token-based billing rules. Higher tiers include more model allowance and sandbox runtime.
Model allowance is measured at standard API prices, while the purchase price is about 30% of the equivalent model-usage value.
Model allowance · Deducted by token
Input, output, and cache-hit tokens are calculated separately at the corresponding model provider's standard API price, never by request count or task duration.
Cloud sandbox · Actual runtime
Sandbox runtime is tracked independently from model-token allowance and is deducted only while the isolated cloud computer is running.
Billing rules
- · Model usage: Deducted entirely by token, with input, output, and cache hits calculated at the corresponding provider's standard API prices.
- · Cloud operation: Deducted from actual isolated cloud-computer runtime and tracked independently from model allowance.
- · Allowance validity: Added immediately after purchase and valid for one month. Each purchase creates a separate monthly grant, and unused allowance does not roll over after expiry.
- · Renewal: Plans are currently paid manually month by month and do not enable automatic WeChat deductions.
Frequently asked questions
Which plan should I choose?
Start with Starter for occasional searches and small analyses. Choose Researcher for ongoing papers, data analysis, and experiments. Professional fits frequent long tasks and long-running computation. All three use the same billing method.
How is model allowance deducted?
Entirely by token. Input, output, and cache-hit tokens are calculated separately at the corresponding model provider's standard API price, never by request count or elapsed time.
Why does a plan cost less than its model-allowance value?
The discount applies when you purchase a plan. You pay roughly 30% of the equivalent value, while usage is still recorded 1:1 at standard API prices so actual consumption remains transparent.
How is sandbox runtime measured?
It is deducted from the actual time the cloud workspace is running and is independent of model-token charges. A stopped sandbox does not consume runtime, and current plans use the same CPU and memory specifications.
Will I lose research files when allowance runs out?
No. New model calls or cloud operation pause, while your account, workspaces, files, and task history remain intact. Purchase another monthly plan to continue.
Spend the budget on more useful iterations
Start from a real research goal and inspect the agent's sources, code, parameters, and deliverables.