# Limits, Concurrency & Cold Start (https://luxor.ai/docs/runners/limits-and-concurrency)

> For the complete documentation index, see [llms.txt](https://luxor.ai/llms.txt)

Concurrency defaults per plan, cold start latency, queue handling, and billing behavior for Runners.

This page is the single reference for the operational limits of Runners: concurrency, cold start, queue handling, and how usage is billed.

## Cold start

Runners cold-start in approximately **15 seconds on average**, regardless of runner size or concurrency. Every job boots into a fresh VM.

## Concurrency

Your plan sets a single concurrent-session limit, shared by Runner jobs and Sandbox sessions:

| Plan       | Concurrent sessions           |
| ---------- | ----------------------------- |
| Starter    | Up to 5                       |
| Team       | Up to 100                     |
| Enterprise | Custom (unlimited on request) |

Jobs above your concurrency limit queue rather than fail, and start as capacity frees up. Queue time is never billed.

## Size limits

The Starter plan supports runner sizes up to 2 vCPU / 4 GB, or 4 vCPU / 8 GB once a payment method is on file. A Starter job targeting a larger size is blocked rather than queued; switch to a smaller label or upgrade to run it. The Team plan supports sizes up to 16 vCPU / 64 GB, and larger sizes are available on request; see [Pricing](https://luxor.ai/docs/pricing.md) for current plan details.

## Billing behavior

* Usage is billed per second of job runtime; rates are quoted per minute. Billing starts when the job starts running, not when it queues.
* There are no idle charges: VMs are destroyed at job end and there is no standby pool, so nothing accrues overnight or on weekends.
* Jobs are balance-gated: if your workspace has no usable credit, Luxor Agent Cloud rejects new jobs before any paid work starts.
* An optional daily spending limit can cap usage; Luxor Agent Cloud cancels runs beyond it with a "spending limit exceeded" notice.
* The only recurring charges are your plan fee and any active add-ons; see [Pricing](https://luxor.ai/docs/pricing.md).

## Job duration

Runners impose **no job time limit** of their own; GitHub's standard limits (6 hours per job, 35 days per workflow) still apply.

## Cache storage

Cache storage comes from your workspace's storage pool (100 GiB Starter, 1,000 GiB Team), shared with sandbox snapshots and volumes, with LRU eviction; see [Caching](https://luxor.ai/docs/runners/caching.md) for the full behavior.