SI Agent News
DevelopingClaudeGeminiResearch

GitHub to retire six Copilot models by December 2026

GitHub is phasing out several AI models used for code completion and chat, introducing new replacements with specific data retention rules.

Papercut shadow box: picture for the story 'GitHub to retire six Copilot models by December 2026'
AI-made papercut shadow box for this story. No real people or products are shown.
December 14Deprecation deadline

The short version

Why it matters

For developers and research teams, this change means that the underlying logic generating code will shift. Organizations must audit their AI governance to ensure new models meet their specific data privacy and reproducibility standards.

The Deprecation Timeline

GitHub has officially announced that six of its Copilot models are scheduled to be retired on December 14, 2026. This transition impacts a wide range of features within the coding assistant, including Copilot Chat, inline edits, and agent modes. The move is part of a broader effort to streamline the available AI tools for developers.

The retirement notice specifically identifies which models are being phased out to make room for newer technology. While GitHub is removing these older models, it is not requiring users to manually delete them from their systems. Instead, the platform will handle the transition automatically as the deadline approaches.

One notable exception to this deprecation involves the Haiku 4.5 model. According to the announcement, individual Copilot subscribers who are currently on annual plans will not be affected by the Haiku 4.5 retirement. This suggests that different tiers of service may experience different timelines for model updates.

New Model Rollouts

To replace the retiring models, GitHub is introducing several new options including Gemini 4 Hydrogen and Gemini 4 Carbon. Gemini 4 Hydrogen was announced for general availability on November 13, 2026, while Gemini 4 Carbon followed shortly before on November 10. These models are being integrated into the Copilot ecosystem to provide updated capabilities.

The rollout of these new models is not instantaneous for all users. GitHub has stated that the deployment will be gradual, meaning that different users may see the new models at different times. For Business and Enterprise customers, the availability of these models is subject to specific administrator controls within their organizations.

Because the release is gradual, the mere announcement of a new model does not mean every institutional user will see it immediately. Administrators must check their specific model policies to see which versions are active. This distinction between default availability and explicit policy configuration is a key detail for large-scale deployments.

Data Retention and Policies

A critical component of this transition involves how data is handled by the different models. For example, the Claude Fable 5.2 model, released on November 9, has a model policy that is turned off by default. This means that Anthropic retains prompts and outputs for the purpose of safety classifiers unless specific actions are taken.

Eligible enterprise customers have the option to request a temporary zero-data-retention exemption for these types of models. This exemption is available through the end of 2026, but it is not automatic. It requires a separate administrator enablement step to be completed before the data can be protected in this manner.

For institutions that prioritize privacy, the name of a replacement model does not automatically confirm the data handling rules. Teams must verify whether a specific model is included in GitHub’s data-retention agreement. This makes the transition a matter of governance as much as it is a technical update to the coding tools.

Impact on Research Teams

For university research software teams, these changes mean that the underlying engine for their coding assistant is shifting. When a model is retired and replaced, the way code is generated may change. Researchers may need to record which model was used during specific development phases to maintain a clear history of software changes.

Testing the outputs of replacement models is important for maintaining computational reproducibility. The retirement notice does not provide any evidence that the new models will produce the same results as the old ones. Therefore, researchers should verify that the new models do not alter the behavior of the software they are developing.

University IT teams and research offices must view this as an operational deadline for their AI governance frameworks. They need to make three separate decisions: which replacement model to use, which model policies to enable, and whether to approve a specific model for research use. The December deadline makes these administrative tasks a priority for the coming months.

What to watch

Try it

Check your organization's GitHub Copilot model policies to see if Gemini 4 Hydrogen or Carbon are currently enabled.

Sources and links

Liked it? Say why

Add a line for other readers.

No account needed. We never show your email. Comments with links, ads or abuse are held and not shown. Your comment shows after a quick check.

Listen to the AI Wire headlines

Listen to the AI Wire

Listen to our daily AI shows every morning: on your commute, in the shower, wherever you are

Listen to the daily shows

First reported by CASRAI · Read the original. Written by the SI Agent News Desk from the reporting above.Written with AI from the source above. Every number is checked against the source. Spot a mistake? Corrections.

More AI news