GitHub Retires Six Copilot Models on Dec 14, and Claude Code Makes Sandboxed Bash Ask First

At a glance

• Six Copilot models will leave every Copilot surface, so check which ones your team still picks by default.
• A new Gemini model is now generally available in Copilot.
• Claude Code now asks before running sandboxed shell commands unless you switch the old behavior back on.
• Copilot CLI has a new stable build that sizes context per model and lets teams install plugins from a branch or tag.

GitHub posted the model deprecation notice at about 3:40 PM ET on Nov 13, roughly two hours after the Gemini 4 Hydrogen launch. Claude Code v2.1.325 reached GitHub at about 3:40 PM ET the same afternoon, and Copilot CLI v1.0.112 followed at about 6:59 PM ET. The Cursor changelog is unchanged since Nov 6, and the Cursor CLI changelog is unchanged since Oct 22. Codex, Gemini CLI and Qwen Code have no new stable release since Nov 13.

Treat today as a check your defaults day: find out which Copilot models your team still selects before Dec 14, read the Claude Code sandbox change before you update a machine that runs unattended jobs, and refresh Copilot CLI on your own machine once the first two are done.

Top Stories

1. GitHub will retire six Copilot models on Dec 14

GitHub's changelog says Claude Fable 5, Claude Haiku 4.5, Gemini 3.8 Flash, Gemini 4 Argon, Grok 4.6 and Kimi K3 will be deprecated on Dec 14, 2026 across all Copilot experiences, including Copilot Chat, inline edits, ask and agent modes, and code completions. That is 30 days from today. GitHub's table names Claude Fable 5.2 as the suggested alternative to Claude Fable 5, Claude Haiku 5.5 for Claude Haiku 4.5, Gemini 4 Hydrogen for Gemini 3.8 Flash, Gemini 4 Carbon for Gemini 4 Argon, and Grok 4.7 for Grok 4.6. Kimi K3 is listed as to be announced. Claude Haiku 4.5 stays available to individual Copilot subscribers on annual plans.

The same notice says that under default model enablement, the suggested alternatives are switched on automatically for Copilot Business and Copilot Enterprise unless an administrator turned off the global default or disabled the model.

Earlier on Nov 13, GitHub made Gemini 4 Hydrogen generally available in Copilot. The deprecation table lists it as the suggested alternative to Gemini 3.8 Flash.

Practical dev impact: If a workflow, prompt file or team habit names one of the six models, move it before Dec 14. Admins who turned off the global default should enable the alternatives through model policies in Copilot settings, because nothing will switch for them automatically.

2. Claude Code v2.1.325 sends sandboxed Bash commands through the permission flow

The biggest change in v2.1.325 is a default. The notes say sandboxed Bash commands now follow the normal permission flow by default, and you set sandbox.autoAllowBashIfSandboxed: true to keep running them without prompts. If managed settings turn the sandbox on, that setting belongs in your highest ranking managed source.

Several other changes point the same way. Read only git commands now ask first in a repository that is not the project's or one you trusted, and the notes say to name that repository with --add-dir to run them without prompts. Scheduled tasks and /loop now ask before running a skill that declares allowed-tools or hooks, unless you add Skill(name) to permissions.allow. The release also fixes a security issue in how zstd compressed responses are decoded and closes a gap where some compound Bash commands could run without asking.

npm latest and next for @anthropic-ai/claude-code both point to v2.1.325, while stable is still v2.1.311.

Practical dev impact: A scheduled task or /loop run that used to work through sandboxed shell commands may now stop and wait for you. Decide on purpose whether that machine should keep prompting, and only add the auto allow setting where you trust the sandbox to do the guarding.

3. Copilot CLI v1.0.112 sizes context per model and pins plugins by branch or tag

Copilot CLI v1.0.112 is now the stable build on npm latest. The notes say the CLI uses each model's recommended context window by default while keeping any context choice you set yourself. Team plugins can now install from marketplaces on a specific branch or tag, or from a repository subfolder, and subagents now respect the session credit limit set with --max-ai-credits.

Earlier on Nov 13, v1.0.111 added regular expression support to enterprise managed Shell, Bash and PowerShell permission rules. The npm prerelease tag points to v1.0.112-0, which belongs on a test machine only.

Practical dev impact: If you cap spend with --max-ai-credits, expect subagent runs to stop at that cap now. Try branch or tag plugin pins on one machine first.

Practical Impact Analysis

Our read: the Copilot deprecation is the item with a real deadline and the widest reach, because it touches every Copilot surface and every plan. Teams on default enablement will see the alternatives appear on their own, so the work is mostly finding saved prompts and scripts that still name a retiring model.

The Claude Code change is smaller in reach but more likely to surprise someone on Monday, because an unattended job that ran sandboxed commands without prompts can now sit waiting for an answer. Read the notes before updating a build box or a machine that runs /loop, then decide whether that machine should prompt. Copilot CLI v1.0.112 is a short, useful refresh, with the credit limit change as a guard for subagent users. Elsewhere, Codex npm latest is still v0.169.0 with only an alpha moving, Gemini CLI latest is still v0.68.0, and Qwen Code latest is still v0.27.3. If you have one hour, list the models your team uses and read the Claude Code sandbox lines.

Tutorial

1. Open a test shell. The active commands below only print registry tags, versions and file names.
2. Run step 2 from the root of a project where you use Claude Code.
3. Uncomment an install line in step 3 only if you want the pinned builds.
4. Step 4 counts the days left before the Copilot deprecation from your own clock.

# We have not run this script end to end on every platform. It prints and checks, and changes no files.
# Run it in a test shell. Nothing is written to your shell profile.

# 1. Snapshot at 6:01 AM ET Sat Nov 14 2026, npm registry channel:
#    @anthropic-ai/claude-code latest v2.1.325, stable v2.1.311
#    @github/copilot latest v1.0.112, prerelease v1.0.112-0
npm view @anthropic-ai/claude-code dist-tags.latest dist-tags.stable
npm view @github/copilot dist-tags.latest dist-tags.prerelease
if command -v claude >/dev/null; then claude --version; else echo "no claude on PATH"; fi
if command -v copilot >/dev/null; then copilot --version; else echo "no copilot on PATH"; fi

# 2. Claude Code: does this project already set the sandbox auto allow key?
#    v2.1.325 makes sandboxed Bash ask first unless the key is true.
for f in .claude/settings.json .claude/settings.local.json; do
  if [ -f "$f" ]; then
    if grep -l "autoAllowBashIfSandboxed" "$f"; then
      :
    else
      echo "$f has no auto allow key"
    fi
  fi
done

# 3. Optional installs only. Pinned at 6:01 AM ET Sat Nov 14 2026, npm registry channel:
# npm install -g @anthropic-ai/claude-code@2.1.325
# npm install -g @github/copilot@1.0.112
#    Test machine only, npm prerelease channel, not for production:
# npm install -g @github/copilot@1.0.112-0

# 4. Days left before the Dec 14, 2026 Copilot model deprecation, from your clock.
python3 -c 'import datetime; print((datetime.date(2026, 12, 14) - datetime.date.today()).days, "days left")'

echo "Done. Nothing was installed or changed on disk."

If npm view shows newer tags, refresh before you pin.

Sources:

• Upcoming deprecation of selected GitHub Copilot models in December
• Gemini 4 Hydrogen in GitHub Copilot
• Claude Code v2.1.325 release
• Copilot CLI v1.0.112 release
• Copilot CLI v1.0.111 release
• Copilot CLI v1.0.112-0 release
• Cursor changelog
• Cursor CLI changelog
• GitHub Changelog

Recommended AI prompt

Copy this paragraph into ChatGPT, Claude, Gemini, Grok, or whatever AI you use.