# Google Colab

> Source: https://aiwiki.ai/wiki/google_colab
> Updated: 2026-10-10
> Fact-checked: 2026-10-10
> Categories: Data Science, Developer Tools, Google
> License: CC BY 4.0 (https://creativecommons.org/licenses/by/4.0/) - attribute to "AI Wiki (aiwiki.ai)"
> Cite as: AI Wiki. "Google Colab." aiwiki.ai, 10 Oct 2026. https://aiwiki.ai/wiki/google_colab
> From AI Wiki (https://aiwiki.ai), the free encyclopedia of artificial intelligence. Reuse freely with attribution.

Google Colab (Colaboratory) is [Google](https://aiwiki.ai/wiki/google)'s hosted Jupyter notebook service for browser-based [Python](https://aiwiki.ai/wiki/python), [machine learning](https://aiwiki.ai/wiki/machine_learning), and data analysis, with [GPU](https://aiwiki.ai/wiki/gpu) and [TPU](https://aiwiki.ai/wiki/tpu) compute.[1]

## Notebooks and runtimes

Notebooks are stored in Drive, can load from [GitHub](https://aiwiki.ai/wiki/github), and download as `.ipynb`. Sharing includes text, code, saved outputs, and comments, but excludes the runtime's installed libraries and files.[1]

Hosted code runs in a private virtual machine that is deleted after inactivity or its maximum lifetime. Hardware availability and usage limits vary, including on paid plans. Premium GPU access is not guaranteed; exhausting a paid compute-unit balance restores free-tier restrictions.[1]

## Paid access and Google AI plans

Colab offers Pro, Pro+, and Pay As You Go. Eligible paid Google AI plans supply monthly compute units; benefits vary by plan and region. Colab-native and Google AI subscriptions can coexist, adding units to one balance on different schedules.[1]

| Google AI plan | Colab compute-unit allowance |
|---|---|
| Pro | 200 CCUs[2] |
| Ultra | 1,000 or 2,000 CCUs, depending on the plan[3] |

The Google One Colab benefit is for users aged 18 or older and family-plan managers, is available on the web, and excludes trial memberships.[2][3] Ultra also includes longer background execution and premium GPU access.[3]

Google's Gemma account advertised access to [NVIDIA A100](https://aiwiki.ai/wiki/nvidia_a100) GPUs with 80 GB of memory through Google AI Pro, and [NVIDIA H100](https://aiwiki.ai/wiki/nvidia_h100) GPUs through Ultra.[4] These advertised hardware options remain subject to Colab's availability and resource limits.[1]

## AI assistance

Google expanded its AI-first Colab interface to general availability on June 24, 2025. The interface combines conversational coding with a Data Science Agent that prepares an analysis plan, executes code, examines results, and presents findings. Users can give feedback during execution.[5]

The assistant can explain libraries, generate code, propose debugging changes in a diff view, and refactor code in response to a natural-language request. It also supports analytical workflows that prepare data, train models, evaluate results, and produce charts.[5]

## Terminal workflows

Google introduced the Colab command-line interface on June 5, 2026, allowing terminal users and agents to work with remote Colab runtimes.[6] The CLI supports Linux and macOS. Its repository documents Python-script and notebook execution, remote file management, and export of session logs.[7]

| Command | Purpose |
|---|---|
| `colab new --gpu A100` | Request an A100 runtime[7] |
| `colab exec -f train.py` | Execute a local Python file remotely[7] |
| `colab download REMOTE LOCAL` | Retrieve a remote file[7] |
| `colab stop` | Release the runtime[7] |

The CLI sends a local script's contents to the remote kernel when using `exec -f`; it does not require a separate upload first. `colab run` combines provisioning, execution, file retrieval, and runtime cleanup in one job.[7]

## Local runtime connections

Colab can connect to Jupyter or Google's Colab Docker runtime on a user's own machine. This lets the notebook execute code using local hardware, but that code can read, modify, and delete local files. Sharing the notebook does not share the local runtime; other users normally receive a hosted runtime.[8]

Google labels its Docker runtime image for demonstration use rather than production because its dependencies can be outdated and vulnerabilities may remain untriaged. Code-cell outputs are saved to Drive by default, with a notebook setting to omit them.[8]

## References

1. Google Colaboratory. [Frequently Asked Questions](https://research.google.com/colaboratory/faq.html). Accessed October 10, 2026.
2. Google One Help. [Use Google AI Pro benefits](https://support.google.com/googleone/answer/14534406?hl=en), Use Google Colab. Accessed October 10, 2026.
3. Google One Help. [Use Google AI Ultra benefits](https://support.google.com/googleone/answer/16286513?hl=en-419), Use Google Colab. Accessed October 10, 2026.
4. Google Gemma. [Google AI Pro and Ultra Colab hardware announcement](https://x.com/googlegemma/status/2108652628590944540). X. Accessed October 10, 2026.
5. Nick Sharma, Jeffrey Mew, and Ilai Soloducho. [Supercharge your notebooks: The new AI-first Google Colab is now available to everyone](https://developers.googleblog.com/en/new-ai-first-google-colab-now-available-to-everyone/). Google Developers Blog. June 24, 2025.
6. Spencer Shumway, Tyler Pirtle, and Seth Troisi. [Introducing the Google Colab CLI](https://developers.googleblog.com/introducing-the-google-colab-cli/). Google Developers Blog. June 5, 2026.
7. Google Colab. [Google Colab CLI repository](https://github.com/googlecolab/google-colab-cli), README and command index. Accessed October 10, 2026.
8. Google Colaboratory. [Local runtimes](https://research.google.com/colaboratory/local-runtimes.html). Accessed October 10, 2026.