Skip to content

GitHub Copilot adds Claude Fable 5.2 and Gemini 4 Carbon

GitHub's weekly Copilot roundup brings two new models, phone handoff for cloud and Codespace sessions, GitHub Enterprise Server support in the Copilot app, and an experimental session inbox in VS Code 1.146.

Draftship Editorial3 min read

What landed this week

On November 16, 2026, GitHub published its weekly roundup of Copilot changes for the week of November 9. The GitHub Copilot new models in this batch are Claude Fable 5.2 and Gemini 4 Carbon, and they arrive alongside GitHub Enterprise Server repositories in the standalone Copilot app and a session inbox in VS Code. The changelog post covers Copilot itself, the Copilot app and the VS Code 1.146 release in one entry.

It is a roundup rather than a single launch. Some items are generally available, others are labelled experimental or preview, and the post gives no benchmarks or performance figures for any of it.

Two new models, and one admin switch

Both models are listed for Copilot Pro+, Max, Business and Enterprise users. Claude Fable 5.2 carries two conditions the source spells out: it requires data retention by default, and Copilot Business and Enterprise administrators must turn on its policy, which is off by default. Gemini 4 Carbon has no such caveat in the post.

That switch matters for teams on business plans. A paid seat is not enough on its own; an administrator has to enable the policy first.

Copilot app: phone handoff, MCP sign-in, enterprise repos

  • Cloud and Codespace sessions can continue on a phone through a new Open in menu with a GitHub Mobile option that shows a QR code.
  • When an MCP server such as Slack needs a fresh sign-in, the conversation offers the matching connector, connects it in place and resumes the task without leaving the chat.
  • Repositories, issues and pull requests from GitHub Enterprise Server now work in the Copilot app, generally available. A GitHub.com or GHE.com account is still required, and cloud sessions are not available for GHES.
  • Project automations can be scheduled to the minute, or with a custom cron expression such as weekdays at 9:30.
  • Voice dictation covers French, Spanish and other languages after downloading the multilingual voice model in settings, with an Auto option or a specific language.
  • Issues and pull requests can be filtered by custom fields, such as a GitHub Project's Size, from the Add filter menu.
  • High contrast can be combined with red, green or blue-yellow color filters, with a preview before applying.

VS Code 1.146: agents that compare models

  • With Compare Models enabled and agent tools switched on in settings, agents can find available models, launch comparisons and read the results to summarize findings. GitHub marks this experimental.
  • Inbox mode in the Agents window's sessions list tracks sessions across all workspaces, putting the ones that need your attention first, with options to dismiss, hold or snooze the rest. Also experimental.
  • A chat can be moved out of the Agents window into its own window and arranged alongside its editors and terminals. Chat windows are in preview.
  • Extensions can be found, installed and uninstalled inside the Agents window, with a warning for unsupported features. Experimental.
  • GitHub asks users to move to the Copilot harness before the Local harness is removed in February 2027. The Copilot harness is already the default for new sessions.

Limits and open questions

There is no pricing detail for either model, no explanation of what data retention means for a given organization, and no date for when the experimental VS Code features become stable. Related posts on the same page reference Copilot model deprecations on November 16, 2026 and in December, but the captured text does not name the affected models.

What this suggests for builders

Inference, not confirmed by the source: if agents can run model comparisons themselves and summarize the results, evaluation likely becomes a routine step inside the editor rather than a manual experiment. The admin policy on Claude Fable 5.2 also suggests that in Business and Enterprise setups, availability depends as much on organizational settings as on the plan someone holds.

Choosing a model for an app you are building is a different decision from choosing one for your editor, and Draftship's guide to using the AI chat and choosing a model covers how that works there. Draftship is not affiliated with GitHub and does not claim support for any model named in this announcement.

Primary sources

This article was drafted with AI assistance from the primary sources listed below and checked automatically against them. An editor hasn't reviewed it yet.

AI coding agents

opencode v2.3.2 fixes GPT-6.2 prompt caching billing

opencode v2.3.2 sends the system prompt as a leading developer message on every Responses route so GPT-6.2 caches it, cutting uncached input per step from about 14,000 tokens to 3 on one Copilot model.