Google Colab · Rate Limits
Google Colab Rate Limits
Colab does not expose a request-rate API; instead it enforces session-level limits on runtime duration, idle disconnect, concurrent sessions, and GPU/TPU availability. Resource access is governed by your subscription tier and remaining compute units. Colab Enterprise inherits the standard Vertex AI / Notebooks API quotas.
7 Limits
NotebooksMachine LearningGoogle CloudRate Limiting
Limits
Maximum runtime duration (Free) session
12
Free notebooks disconnect at 12 hours of continuous runtime; idle disconnects much earlier.
Idle timeout (Free) session
90
Approximate; Colab disconnects idle Free sessions sooner than paid tiers.
Maximum runtime duration (Pro) session
24
Maximum runtime duration (Pro+) session
24
Pro+ also supports persistent background execution beyond the active session window.
Concurrent sessions per user user
1
Free and Pro typically support 1 active high-resource session; Pro+ allows additional concurrent sessions.
GPU/TPU availability user
see availability
Premium GPUs (A100/H100/L4) are subject to availability and prioritized for paid tiers.
Compute unit consumption subscription
see live runtime meter
Compute units are consumed per hour of attached runtime; rate varies with GPU/TPU class.
Policies
Session lifecycle
Sessions terminate on idle disconnect, runtime cap, or browser close. Save progress to Drive or GitHub frequently.
Resource gating by tier
Free tier is best-effort and may queue or refuse high-end GPUs during peak load. Paid tiers receive priority but are still capacity-limited at the global pool.
Compute unit metering
Pro/Pro+ subscribers consume from a monthly compute-unit bucket; PAYG users consume from a 90-day bucket. Once exhausted, runtimes downgrade to Free behavior.
Enterprise quotas
Colab Enterprise (Vertex AI Workbench) follows standard Vertex AI / Compute Engine quotas enforced at the GCP project level; raisable via the Cloud Console quotas page.