Skip to content

Factories > Plans and support

Usage and billing

Track Warp usage, understand inference, compute, and platform costs, and distinguish accounts billed in dollars from Enterprise contracts using credits.

Open in ChatGPT ↗
Ask ChatGPT about this page
Open in Claude ↗
Ask Claude about this page
Copied!

Warp usage pays for model calls, hosted compute, and platform services. For Warp Terminal and local agents, open Settings > Billing and usage in the Warp app. For Warp Factories, open Settings > Billing and usage in the Warp Factories web app.

Your plan determines the unit Warp uses to charge usage:

  • Dollar billing - Your usage balance is an amount in US dollars. Usage charges draw from that balance.
  • Credit billing - Usage is measured and charged in credits. Enterprise contracts that use credits retain their rates and terms.

Changing the display unit doesn’t change your billing terms or charges. If you’re unsure which unit your account bills in, check your billing terms or contact billing@warp.dev.

Usage has three charge components:

  • Inference - Model calls provided by Warp. If you use your own API key or inference endpoint, your provider bills that inference separately.
  • Compute - The sandbox used by an agent on Warp-hosted compute. Local runs and self-hosted compute don’t incur this charge.
  • Platform - Billable agent time for cloud runs and eligible local runs using customer-supplied inference. See platform usage for product-specific eligibility and metering.

For paid self-serve plans billed in dollars, Warp-provided inference is priced at public API rates. Compute and platform charges are separate. Enterprise terms follow your contract.

A credit is a usage unit, not a token or a fixed number of prompts.

  • Account usage - Check your balance and reset date in Settings > Billing and usage.
  • Turn usage - Expand the usage amount below an agent response. Where detailed usage is available, expand a model to see input, output, cache-read, and cache-write tokens, plus web searches and their costs.
  • Conversation usage - Open the conversation’s usage summary to see its total. View account usage opens Billing and usage.

The CLI has its own usage display.

Recorded dollar totals represent Warp usage charges, not the total price of your subscription or usage purchases. Older usage can have estimated costs, credit-only details, or no breakdown. An unavailable amount does not mean the run was free. Bills from your own inference provider are not fully represented in Warp’s totals.

Build, Max, and Business subscriptions for Warp Terminal and local agents allocate included usage per seat. Included usage resets monthly, including on annual subscriptions, and does not roll over. Check Settings > Billing and usage in the Warp app for your allowance rather than assuming it matches a new subscription.

After a member’s allowance runs out, eligible runs draw from purchased usage. The Free plan for Warp Terminal and local agents supports usage purchases without subscribing, or personal inference.

The Scale plan for Warp Factories includes a shared monthly allowance that does not roll over. The Start plan for Warp Factories uses purchased usage without a subscription. Factory runs require shared team or workspace funding, not a member’s included personal allowance.

In the Warp Factories web app, open Settings > Billing and usage to check shared balances. After shared included usage runs out, factory runs use eligible purchased usage. See funding factory runs for covered charges, exhaustion behavior, and Enterprise contract terms.

Purchased usage rolls over until its expiration and is shared across the team. Eligible shared purchases are spent before member-specific purchases usable on the same team. Dedicated cloud allowances can be used before the general balance. Enterprise pool sizes and allocation rules follow your contract.

Generate and AI Autofill in Workflows also use inference. Regular shell commands and non-AI terminal features do not consume agent usage.

Inference cost depends on the model’s rates and the work required to complete your request:

  • Model choice - Models have different input, output, and cache prices. A lower-cost model can reduce spending without reducing the number of tokens.
  • Input and output - Your prompt, conversation history, attached context, and generated response contribute to token usage.
  • Caching - Cached input can cost less than new input. Cache-read and cache-write tokens are reported separately.
  • Task complexity - A task can require several model calls, including calls made while using tools or summarizing a long conversation.
  • Provider tools - Features such as web search can add charges beyond token costs.

Two similar prompts can use different amounts. Use the reported breakdown to compare tasks rather than treating a prompt as a fixed-price unit. See using tokens efficiently for ways to reduce usage.

Cloud runs on Warp-hosted compute incur compute charges, whether started from the Warp app, an integration, the CLI, or the Warp Platform API. Compute cost is the sandbox’s rate multiplied by its metered lifetime, not its CPU utilization or billable agent hours.

Local runs, CI jobs on your own runners, and self-hosted workers do not incur Warp-hosted compute charges. Platform charges can still apply.

The rates below apply to Free, Build, Max, and Business accounts for Warp Terminal and local agents billed in dollars, with no additional compute markup. They are configured compute rates, not a live lookup of a provider’s invoice. Enterprise margins, discounts, and credit conversions follow your contract. Usage purchase premiums affect the amount you pay to buy usage, not these metered rates.

Match your runner’s operating system, architecture, and instance shape to the rates below. Available shapes and default sizes depend on your plan.

Linux x86-64 and aarch64 use the same rate. Windows supports x86-64 and costs twice the Linux rate for the same shape.

To calculate the per-minute rate, take the greater of the vCPU count and half the memory in GB. Multiply that number by $0.0015 for Linux or $0.003 for Windows. For example, 4 vCPUs with 16 GB costs $0.012 per minute on Linux.

Example shapes:

vCPUsMemoryLinux per minuteLinux per hourWindows per minuteWindows per hour
24 GB$0.003$0.18$0.006$0.36
48 GB$0.006$0.36$0.012$0.72
816 GB$0.012$0.72$0.024$1.44
1632 GB$0.024$1.44$0.048$2.88

macOS runners use Apple Silicon and have fixed rates for each supported shape:

vCPUsMemoryPer minutePer hour
47 GB$0.075$4.50
614 GB$0.09$5.40
814 GB$0.135$8.10
1228 GB$0.18$10.80
1256 GB$0.27$16.20

For a newly created sandbox, compute is metered from creation to destruction. This includes setup, command execution, and idle time while the sandbox remains running. If a run receives a prewarmed sandbox, that run’s metering starts when the sandbox is assigned to it.

Elapsed time is charged proportionally, including partial minutes; it isn’t rounded up to whole minutes or hours. Compute can continue after the agent stops responding if the sandbox stays available. Compute charges are recorded when the sandbox is finalized.

Cloud runs, including factory runs and third-party harnesses, incur platform usage for billable agent time. For Warp Terminal and local agents, Business and Enterprise plans also incur platform charges when local runs use customer-supplied inference. The same local-run rule applies to work billed under a team’s Scale plan for Warp Factories. See product-specific platform billing for other billing contexts and Enterprise contract terms.

See platform usage for billable time, exceptions, and the distinction between agent hours and run time.

For a Warp Agent run on a self-serve plan billed in dollars, suppose a 4 vCPU / 8 GB Linux sandbox exists for 10 minutes, with 6 billable agent minutes and $0.40 of Warp-provided inference:

  • Compute - 10 minutes × $0.006 = $0.06.
  • Platform - 6 minutes × $0.05 = $0.30.
  • Inference - $0.40.

The run consumes $0.76 from the eligible usage balance. Subscription payments, usage purchase premiums, and bills from your own inference provider are separate.

For factory runs, see shared funding requirements. The following rules apply to standalone cloud runs.

Agent API key runs follow the team billing rules for cloud runs.

By default, user-created schedules use the creator’s eligible included and purchased usage. A schedule configured to run as a cloud agent uses shared team usage instead. Enterprise pools and charges follow your contract.

If no usable balance remains, a run can fail with an insufficient credits error.