Google Colab¶

Development line: project:google-colab · thread google-colab-platform-evolution
Last event: 2023-05-19 · 2 dated since 2022-08-31 · Researched: 2026-09-04 · confidence: medium

What it is¶

Google Colab is hosted Jupyter for Python users who need a browser notebook instead of a local Jupyter setup.

• Dedicated virtual machine that executes user code.
• GPU and TPU access when compute resources are available.
• Google Drive and GitHub integration that stores, shares, and loads .ipynb files.
• Gemini integration that adds chat, code completion, and a data science agent for eligible accounts.

The free runtime is not guaranteed, typically lasts at most 12 hours, and compute quotas change over time.

The platform fits tutorials, quick prototypes, and interactive ML tasks, but not continuous or fixed compute workloads.

Development line¶

• 2022-08-31 — Google Colab subscription-policy change was publicly reported. The linked record is unavailable for review; the specific subscription change is unconfirmed and cannot serve as an active rule.
• 2023-05-19 — Google Colab introduced AI coding features. Google announced Codey features—natural language code generation, code completion for eligible paid subscribers, and a chat assistant—starting with US paid accounts.

What changed¶

Google Colab updated compute access policies and its built-in AI assistant.

• 2022-08-31: the linked record is unavailable for review, so the subscription change is unconfirmed and cannot serve as an active rule.
• 2023-05-19 (official announcement dated 2023-05-17): Google announced Codey features, including natural language code generation, code completion for eligible paid users, and a chat assistant, starting with US paid tiers.
• 2026-04-08: the Gemini integration added notebook-saved Custom Instructions and Learn Mode for step-by-step tutoring.
• 2026-09-04: current documentation describes AI-First Colab with Gemini chat, code completion, and a Data Science Agent that proposes a plan and runs it after user review, with access varying by age and region.
• 2026-09-04: compute allocation is explicitly dynamic; neither free nor paid tiers guarantee resources, and limits, VM lifetimes, and accelerator types fluctuate.

The early Codey rollout does not reflect the current Gemini interface, and prior pricing statements do not replace the active subscription page.

Colab remains a single service, but users must evaluate available compute runtimes and AI tools separately.

How to use this¶

After 2023-05-19, practitioners should assess Google Colab as both a hosted notebook environment and an AI-assisted coding environment, while checking the applicable subscription and access policy for their workload.

1. Open Colab, create a new notebook through File, or upload an existing .ipynb file through File > Upload notebook. — https://research.google.com/colaboratory/faq.html
2. Add code and text cells; if the workload requires hardware acceleration, select Runtime > Change runtime type to pick a GPU or TPU. — https://research.google.com/colaboratory/faq.html
3. Save notebooks to Google Drive or load them from GitHub; share notebooks via Share, keeping in mind that the underlying VM and local files do not transfer. — https://research.google.com/colaboratory/faq.html
4. Open the Gemini assistant at the bottom of the notebook, describe the task, and inspect any generated plan before running the code. — https://research.google.com/colaboratory/faq.html
5. To use a local machine, run Jupyter or a supported local runtime, select Connect > Connect to local runtime, and provide the backend URL. — https://research.google.com/colaboratory/local-runtimes.html

Best practices¶

• Treat managed VMs as temporary: install dependencies and download input files directly in the notebook, and persist results outside the runtime. — https://research.google.com/colaboratory/faq.html
• Avoid requesting GPUs or high memory unnecessarily, and close idle tabs to conserve compute quota. — https://research.google.com/colaboratory/faq.html
• Pin library versions explicitly for shared or teaching notebooks; select an older runtime version when necessary, noting that prior runtimes have limited lifespans. — https://research.google.com/colaboratory/runtime-version-faq.html
• Review, test, and validate code generated by Gemini; keep confidential or personal data out of AI prompts. — https://research.google.com/colaboratory/faq.html
• Connect local runtimes only to trusted notebooks with verified code, because local runtimes can execute arbitrary shell commands and read local files. — https://research.google.com/colaboratory/local-runtimes.html

Superseded by this¶

• 2022-08-31: unverified claims about subscription rules are obsolete; current policy defines all quotas and compute resources as dynamic.
• 2023-05-17/2023-05-19: the Codey rollout for US paid accounts is replaced by AI-First Colab using Gemini for eligible users.
• 2025-03-21: launching Colab via GCP Marketplace VMs is deprecated; Google documentation directs users to Colab Enterprise or local runtimes.

Still unknown¶

• The source text from 2022-08-31 is inaccessible, leaving the exact subscription policy change unconfirmed.
• The recorded event date 2023-05-19 differs from the official announcement date of 2023-05-17.
• Compute limits and AI coding belong to the same service, but Google has confirmed no causal link between the policy shifts.
• Exact pricing, compute units, GPU and TPU models, and Gemini availability vary by account, tier, and region, and require checking the subscription interface directly.

Sources¶

Agent brief¶

• Subject: project:google-colab, thread google-colab-platform-evolution, 2 dated events 2022-08-31 → 2023-05-19.
• Practical note: After 2023-05-19, practitioners should assess Google Colab as both a hosted notebook environment and an AI-assisted coding environment, while checking the applicable subscription and access policy for their workload.
• Confidence: medium. Dated supersedes above are the authority for what is obsolete.