Gemini CLI 0.67: New Features, Models Command and Fixes

Q: How do I list available models in Gemini CLI?

Run gemini models or gemini --list-models. Add -o json to get a JSON array with each model's ID, display name, description and input and output token limits, which you can pipe to jq.

Last updated: November 4, 2026

Gemini CLI 0.67, released November 4, 2026, adds a gemini models command for listing models as text or JSON, brings IDE diffs and open-file context into container sandboxes, and fixes cancellation and quota handling. It follows 0.66, which rebuilt terminal rendering to stop flicker.

Contents

What’s new at a glance
The new gemini models command
IDE integration inside container sandboxes
Reliability and security fixes in 0.67
What 0.66 added two weeks earlier
How to upgrade and try the new features
Who should still use Gemini CLI
Frequently asked questions

Gemini CLI is Google’s open-source terminal agent, licensed under Apache 2.0. The v0.67.0 release on GitHub went out as the new stable build on November 4, 2026, one week after v0.66.0 on October 27. Most entries are bug fixes, but two features matter for daily use. This guide covers both and explains where the tool now fits, since Google moved consumer users to another CLI in June.

What’s new at a glance

Area	Change	Version
Model discovery	New gemini models command and –list-models flag, with JSON output	0.67.0
Sandboxed IDE integration	/ide status, open-file context and native diffs work in Docker, Podman and gVisor sandboxes	0.67.0
Cancellation	Tools no longer run after a cancel in non-interactive mode, and auto-nudge stops after a cancelled request	0.67.0
Quota errors	Project spending cap errors are treated as terminal quota errors	0.67.0
Untrusted flags	Flag checks are scoped to external output, and approved flags are remembered	0.67.0
Settings migration	${VAR} placeholders survive deprecated settings migration; Claude hook timeouts convert from seconds to milliseconds	0.67.0
Terminal rendering	Synchronized output, isolated spinner rendering and preserved scrollback	0.66.0
MCP OAuth	Opt-in support for private network endpoints in MCP OAuth flows	0.66.0

The new gemini models command

Before 0.67 there was no scriptable way to ask Gemini CLI which models your account could reach. The pull request that added the feature introduces gemini models (also gemini models list) and a top-level --list-models flag. Both print a readable list by default. With --output-format json or -o json, they return a JSON array in which each entry carries modelId, id, name, displayName, description and contextLimits with inputTokens and outputTokens.

The flag exits before the sandbox or terminal UI starts, and it skips the non-interactive auth refresh, so credential log lines do not end up in the output. That makes it safe to pipe, for example gemini --list-models -o json | jq -r '.[].id'. A practical use is a CI check that fails fast when a model ID your scripts pass with --model is no longer available to the project.

IDE integration inside container sandboxes

Gemini CLI can run its tools inside a container when you set GEMINI_SANDBOX to docker, podman or runsc (gVisor). Until this release, IDE features such as /ide status, /ide enable, open-file context and native diffs did not work in that mode, because the IDE companion extension listens only on the host and requires a per-window token that is never passed into the sandbox.

Version 0.67 adds a bridge: the launcher on the host relays the IDE session into the container over docker exec or podman exec and a Unix socket. The token stays on the host, and the bridge only allows diff operations on files inside the IDE workspace. Because it does not rely on container networking, it also works under gVisor. The container image needs nothing extra beyond node.

Reliability and security fixes in 0.67

The rest of the release tightens behavior that matters in scripts and long sessions:

In non-interactive mode, the CLI now handles termination signals and will not start a tool after a request is cancelled.
A Google Cloud project spending cap error is now classified as a terminal quota error, so runs stop instead of retrying.
Every model turn that contains function calls now receives a thought signature.
Atomic file writes respect read-only files and hard links, and CRLF line endings are normalized before diffs.
Relative file paths resolve against the active IDE workspace root, and IDE settings and GEMINI.md files are no longer added to open-file context.
ACP permission requests now include the MCP server and tool names, and custom headers are split according to RFC 9110 token rules.

What 0.66 added two weeks earlier

The official release notes list three highlights for v0.66.0. Terminal rendering now uses synchronized output with isolated spinner rendering, which removes buffer flicker and keeps scrollback intact when the view refreshes. The scheduler emits tool call groups directly, and termination signals are forwarded from the bootstrap process to the child process. For security, MCP OAuth flows can opt in to private network endpoints, and shell and git commands keep your ~/.gitconfig while interactive SSH and credential prompts are suppressed so runs do not hang. Version 0.66 also added a keys array in keybindings.json and a banner noting that Gemini CLI is not optimized for Gemini 4.

How to upgrade and try the new features

Update the stable build with npm install -g @google/gemini-cli@latest, or use brew upgrade gemini-cli if you installed with Homebrew.
Run gemini --version and confirm it reports 0.67.0. Version 0.66 added a check for an older binary shadowing the update on your PATH.
If you use a Code Assist license, set GOOGLE_CLOUD_PROJECT to your project ID before running gemini. Otherwise set GEMINI_API_KEY or configure Vertex AI.
Run gemini models -o json to see the models and token limits your credentials can use.
To test sandboxed IDE support, start the CLI from your IDE’s integrated terminal with GEMINI_SANDBOX=docker and run /ide status.
To try upcoming changes, install @google/gemini-cli@preview; v0.68.0-preview.0 was published the same day.

Who should still use Gemini CLI

The project README now opens with a status notice. Gemini CLI is designed for organizations with Gemini Code Assist Standard or Enterprise licenses and Code Assist for GitHub on Google Cloud, plus legacy paid Gemini API and Vertex AI keys. Google’s May 19, 2026 transition post says Gemini CLI stopped serving Google AI Pro, Ultra and free individual users on June 18, 2026, and that those users should move to Antigravity CLI. Our guide to whether Gemini CLI is free and article on whether Gemini CLI trains on your code cover those account questions.

The README also says Gemini CLI is not tuned for the Gemini 4 series and should not be used to benchmark Gemini models. If you want Gemini 4 Argon in a terminal agent, Google recommends the Antigravity CLI. For direct API calls from scripts and other agents, the Gemini API CLI is the stateless option. Among competing terminal agents, Codex CLI 0.166 is a useful comparison point for teams outside Google Cloud.

Frequently asked questions

What is new in Gemini CLI 0.67?

Version 0.67.0, released November 4, 2026, adds a gemini models command and a –list-models flag for model discovery, makes IDE integration work inside Docker, Podman and gVisor sandboxes, and fixes cancellation, quota error and settings migration bugs.

How do I list available models in Gemini CLI?

Run gemini models or gemini –list-models. Add -o json to get a JSON array with each model’s ID, display name, description and input and output token limits, which you can pipe to jq.

Is Gemini CLI still free?

No. According to the project README, free individual access and Google AI Pro and Ultra accounts moved to Antigravity CLI on June 18, 2026. Gemini CLI is now aimed at Gemini Code Assist Standard and Enterprise, Code Assist for GitHub on Google Cloud, and paid API or Vertex AI keys.

Should I use Gemini CLI with Gemini 4 models?

Google says Gemini CLI is not tuned for the Gemini 4 series, such as Gemini 4 Argon, and should not be used to benchmark Gemini models. It points Gemini 4 users to Antigravity CLI.

How do I update to Gemini CLI 0.67?

Run npm install -g @google/gemini-cli@latest, or upgrade with Homebrew or MacPorts if you installed it that way. Then run gemini –version to confirm the new version is the one on your PATH.

How often does Gemini CLI release?

The README says preview releases are published each week on Tuesdays at 23:59 UTC, stable releases promote the previous week’s preview plus bug fixes, and nightly builds ship every day.

Sources

Image: Google Gemini logo from Wikimedia Commons, public domain (trademarked). Gemini is a trademark of Google LLC. FreeAnswers is not affiliated with or endorsed by Google.

Gemini CLI 0.67: Power-User Guide to the November 2026 Release