Usage

Every plan includes a monthly allowance for metered usage. Settings → Billing shows the live meter and category breakdown.

On a paid plan, Auto on a CPU sandbox is the free floor: the allowance bar does not move at all. It fills only while metered work runs — here, the two cells a GPU is attached for — and stops the moment the GPU comes off.

What is metered

Three categories can draw down the allowance:

• Models: built-in model calls other than Auto and DeepSeek V4 Flash. More capable models and higher reasoning effort generally cost more.
• Sandbox: wall-clock time on a metered runtime. GPUs cost more than CPU, and larger GPUs cost more than smaller ones. The meter runs while the sandbox is up, including idle time before its automatic shutdown, and stops when it pauses or is released.
• Web tools: agent calls that search, fetch, or extract information from the web.

Editing, creating notebook cells, branching, and reading output do not themselves consume model allowance. Code still runs inside a sandbox, so metered sandbox time continues while those cells execute.

The unmetered floor on paid plans

On every paid plan, these are unmetered:

• Auto
• DeepSeek V4 Flash
• CPU runtime

That makes routine agent work on CPU available even after the included allowance is gone. DeepSeek V4 Pro, the open-model tier, Claude/GPT, GPUs, and web tools remain metered.

Free works differently

On Free, Auto and CPU are what the free allowance buys, so their use is metered. The built-in picker is locked to Auto and GPUs are unavailable, leaving web-tool calls as the other metered category.

On Free, the remaining allowance is also an in-run ceiling. A turn that reaches it ends gracefully; later agent and execution starts remain blocked until the monthly reset. You can keep reading, editing, downloading, and exporting work.

Bring your own key and a paired local runner use the provider account or CLI subscription you already have, so their model inference does not consume the Clusy allowance. Sandbox and web-tool usage still follow the normal rules.

Allowance and pay-as-you-go

The allowance resets at the start of each UTC calendar month, independently of a Stripe subscription renewal date. Warnings appear at 75%, 90%, and exhaustion.

On a paid plan, exhausted metered work falls back to the unmetered CPU path unless pay-as-you-go is enabled. Pay-as-you-go continues metered work up to a monthly cap you set.

On a paid plan, a run that begins inside the included allowance is not stopped by crossing it; the next metered start applies the fallback. Free uses the in-run ceiling described above. Pay-as-you-go has its own hard cap and can stop metered work during a run once that cap is reached.

Credits

Two credit pools can appear in Billing, and they are intentionally separate:

• Bonus usage is extra included usage for one UTC calendar month. It is folded into that month's meter, expires at month end, and does not roll over.
• Invoice credit is money applied automatically to a future Stripe invoice, including eligible pay-as-you-go charges. It does not extend the usage meter and does not expire.

Open Credit history in Billing to see when either kind was granted.

Reading the meter

The Billing page shows:

• the percentage used this cycle,
• the split across Models, Sandbox, and Web tools,
• the models and runtimes with the most usage,
• and a projection based on the current pace.

To reduce usage, keep routine work on Auto or DeepSeek V4 Flash and CPU, lower effort for mechanical tasks, and attach a GPU only for cells that need it. A branched experiment running N variants on a GPU uses roughly N times the compute of one variant.