What is it

Google Colab (Colaboratory) is a cloud-hosted Jupyter notebook environment that requires no local setup. The Free tier ships with T4 GPU access and a usable AI coding experience; paid options include Pay As You Go ($11.79 for 100 compute units, $58.99 for 500), Colab Pro+ ($58.99/month with 500 included units, priority premium GPUs, and 24-hour background execution), and Colab Enterprise (consumption-priced, tightly integrated with BigQuery and Vertex AI). Compute units expire after 90 days.

What it can do

Write and run Python in a browser-hosted notebook; mount Google Drive for datasets and checkpoints; access GPUs (T4, L4, A100 40 GB / 80 GB, H100, G4 / RTX PRO 6000) and TPUs depending on plan; use Gemini in Colab to chat about the notebook, generate code, complete code, explain code cells, and explain/fix errors; run the Data Science Agent to auto-generate exploratory data analysis and ML plans from a prompt — backed by sklearn, BigQuery ML, BigQuery DataFrames, or PySpark; share notebooks live like a Google Doc.

Who is it for

Students, ML learners, researchers, and data scientists who need GPU/TPU compute without local hardware, and engineering/analytics teams on Colab Enterprise who want notebooks tightly bound to BigQuery, Vertex AI, and Google Cloud IAM.

Key Features

Hosted Jupyter Notebooks in the Browser

A full Jupyter notebook environment with code, markdown, and inline outputs runs entirely in a browser tab — no installation, no environment setup. Notebooks save to Google Drive by default and can be shared with anyone the way a Google Doc is shared.

Free GPU and TPU Tier

The Free plan offers GPU access (typically NVIDIA T4) with no subscription; usage is rate-limited and resources are not guaranteed. The free tier is one of the few public on-ramps where anyone with a Google account can run real GPU-backed ML code.

Premium GPU Lineup with Compute Units

Paid tiers expose faster GPUs including NVIDIA L4, A100 (40 GB and 80 GB), H100, and G4 (NVIDIA RTX PRO 6000, ~96 GB VRAM). GPUs are billed via compute units — 1 unit = $0.10 — with usage rates that scale with the chip (T4 cheapest, H100/G4 the most expensive). V100 has been removed from the lineup.

Pay As You Go and Colab Pro+ Plans

Pay As You Go: $11.79 for 100 units or $58.99 for 500 units, no subscription. Colab Pro+ at $58.99/month bundles 500 compute units, priority access to premium GPUs, and background execution so a running notebook keeps going up to 24 hours even after the browser closes. Compute units expire after 90 days.

Use Cases

Free GPU runs for ML tutorials and small experiments

Students and beginners use the Free plan to run Kaggle starter kernels, fine-tune small transformers, and try out PyTorch/TensorFlow tutorials on a T4 without buying hardware.

Prototyping deep learning models before scaling out

ML engineers prototype on A100 80 GB or H100 sessions, validate the training loop end-to-end, and then graduate the workload to a dedicated cluster or Vertex AI job.

Exploratory data analysis with Data Science Agent

Analysts open a Colab Enterprise notebook attached to a BigQuery dataset and ask the Data Science Agent to "explore this table and summarize anomalies" — DSA writes the plan, generates the code, and runs it inline.

Teaching Python and ML in classrooms

Instructors share a single notebook link that every student can copy; everyone gets a clean environment with the same library versions, no setup support tickets, and free GPU access for assignments.

Pricing plans

Frequently Asked Questions

Discussion

No comments yet. Be the first to start the thread.

Google Colab