How To Update Your AI Tools to the Latest Frontier Models

A new model just dropped and it is not showing up in your coding agent. Ninety percent of the time, this fixes it:

Close every session of the CLI (all terminal tabs and windows).
Update the CLI using the command for your tool below.
Open a fresh terminal and start the agent again.
Run /model — every major coding agent now uses the same slash command.
Pick the new model if it appears. If it does not, force it with the model flag for your tool.
If it is still missing or disabled, stop. New models roll out in stages, and your account, plan, or org policy may not have it yet. Updating again will not change that.

The cheat sheet, then the details for each platform:

Codex CLI: update with codex update, switch with /model, force with codex -m <model-id>
Claude Code: update with claude update, switch with /model, force with claude --model <model-id>
Gemini CLI: update with npm install -g @google/gemini-cli@latest, switch with /model, force with gemini -m <model-id>
GitHub Copilot CLI: update with npm install -g @github/copilot@latest, switch with /model, force with copilot --model <model-id>
Cursor CLI: auto-updates on launch, or run agent update, then switch with /model

Codex CLI: Getting the Latest GPT Models

The current example: OpenAI previewed GPT-5.6 on June 26, 2026 in three tiers — Sol (flagship), Terra (mid-cost), and Luna (fastest and cheapest). During the preview it is limited to the API and Codex for vetted partners, with general availability promised in the coming weeks. Whenever a new GPT model goes GA, this is how you get it.

Check your version and self-update:

[ bash ]

codex --version
codex update

codex update works on release builds. If it is not supported by your install, update the way you installed:

npm:

[ bash ]

npm install -g @openai/codex@latest

Careful: the package is @openai/codex. The unscoped codex package on npm is an unrelated project from 2012.

Homebrew:

[ bash ]

brew upgrade --cask codex

Standalone installer (macOS or Linux) — rerun it to upgrade:

[ bash ]

curl -fsSL https://chatgpt.com/codex/install.sh | sh

Then start codex, and inside the session type:

[ text ]

/model

That lists available models and reasoning levels. To force a specific model for one session:

[ bash ]

codex -m <model-id>

To make it permanent, set it in ~/.codex/config.toml:

[ toml ]

model = "<model-id>"

Use the exact model ID from OpenAI's release notes — the GPT-5.6 tiers will each ship with their own ID, the same way earlier releases did (for example gpt-5.4 or gpt-5.2-codex).

Claude Code: Getting Fable 5 and Future Claude Models

Fable 5 is fully rolled out again after its June export-control pause, so today the normal path just works. The same path applies to whatever Anthropic ships next.

Check your version and update:

[ bash ]

claude --version
claude update

If you installed via npm:

[ bash ]

npm install -g @anthropic-ai/claude-code@latest

Native installer — rerun it to upgrade. macOS, Linux, or WSL:

[ bash ]

curl -fsSL https://claude.ai/install.sh | bash

Windows PowerShell:

[ powershell ]

irm https://claude.ai/install.ps1 | iex

Package managers:

[ bash ]

brew upgrade --cask claude-code       # macOS Homebrew
winget upgrade Anthropic.ClaudeCode   # Windows WinGet
sudo apt update && sudo apt install --only-upgrade claude-code   # Debian/Ubuntu
sudo dnf upgrade claude-code          # Fedora/RHEL

Then start claude and type:

[ text ]

/model

To force a model directly:

[ bash ]

claude --model claude-fable-5

or inside a session:

[ text ]

/model claude-fable-5

If you need a full uninstall-and-reinstall walkthrough for every installer (npm, Homebrew, WinGet, apt, dnf, apk, native), we covered that step by step in How To Get Anthropic Fable 5 Back in Claude Code.

Gemini CLI: Getting the Newest Gemini Models

Gemini CLI updates through npm even if you launch it other ways:

[ bash ]

gemini --version
npm install -g @google/gemini-cli@latest

If npm cache or permission issues block the update:

[ bash ]

npm install -g @google/gemini-cli@latest --force

Homebrew:

[ bash ]

brew upgrade gemini-cli

Here is the part most people miss: Gemini CLI has three release channels, and new Gemini models often land in preview before stable. Stable ships weekly. If a just-announced model is missing from your stable install, try:

[ bash ]

npm install -g @google/gemini-cli@preview

or the daily build:

[ bash ]

npm install -g @google/gemini-cli@nightly

Then start gemini and type:

[ text ]

/model

That opens a model picker dialog. There is also an Auto option that routes to the right model by task. To force a model at launch:

[ bash ]

gemini -m <model-id>

GitHub Copilot CLI: Switching Between Claude and GPT Models

Copilot CLI is the multi-vendor one — it offers Claude and GPT models side by side, so new models from both companies show up here.

Update:

[ bash ]

npm install -g @github/copilot@latest

Then start copilot and type:

[ text ]

/model

Your selected model is displayed above the input box. To force one at launch:

[ bash ]

copilot --model <model-id>

or set the COPILOT_MODEL environment variable.

One Copilot-specific catch: model availability is controlled by GitHub policy, not just your CLI version. On organization and enterprise plans, an admin may need to enable a newly released model in Copilot settings on github.com before it appears in your picker. If your CLI is current and the model still is not listed, that is the first thing to check.

Cursor CLI: Mostly Automatic

Cursor CLI tries to auto-update every time it starts, so you usually already have the latest version. To force it:

[ bash ]

agent update

Note the command is agent — Cursor renamed it from cursor-agent, though the old symlink may still exist on older installs. Check what you are running:

[ bash ]

agent --version

If you need to reinstall, macOS, Linux, or WSL:

[ bash ]

curl https://cursor.com/install -fsS | bash

Windows PowerShell:

[ powershell ]

irm 'https://cursor.com/install?win32=true' | iex

Then inside a session:

[ text ]

/model

Cursor enables new models server-side, so they typically appear in the picker without any action from you once Cursor turns them on.

If the Model Still Does Not Show Up

Every one of these tools fails the same five ways, and none of them are fixed by reinstalling in a loop:

Your CLI is too old. The local app has to know the model exists. That is what the update step fixes — always do it first.
The rollout has not reached you. Vendors ship new models in stages by account, plan, and region. GPT-5.6 is the current example: previewed to vetted partners first, GA later. Waiting a day beats reinstalling five times.
Your plan does not include it. Flagship models are often gated to paid or higher tiers at launch.
Your org disabled it. Enterprise workspace policies can hide or disable models in Claude Code, Copilot, and Cursor alike. Ask your admin before you debug your terminal.
You are authenticated the wrong way. API-key auth and subscription auth can have different model lists. If you switched auth methods recently, check the other one.

And one thing not to do: do not change your system clock, delete your config directories, or purge credentials to try to force a model. Availability is decided server-side. Locally you only control two things — how new your CLI is, and which model you ask for.

The Pattern To Remember

Model releases now land every few weeks — GPT-5.6 Sol, Terra, and Luna are going GA at OpenAI, Fable 5 just finished its return rollout at Anthropic, and Google pushes new Gemini models through the CLI's preview channel weekly. The tools have converged enough that one habit covers all of them:

[ text ]

close sessions → update the CLI → fresh terminal → /model → force flag → wait

Bookmark this page, and the next time a launch announcement hits your feed, you will be selecting the new model before most people have finished asking why it is not showing up.