Gemini CLI: Gemma 4 Now Enabled by Default via Gemini API

Gemma 4 via Gemini API: Now On by Default

With the release of v0.42.0, Gemini CLI has promoted Gemma 4 model access to a default-enabled feature for all users connecting through the Gemini API. The change is straightforward but meaningful: the experimental.gemma configuration flag, which previously defaulted to false and required a manual settings change, now defaults to true across all platforms.

What Changed

Previously, accessing Gemma 4 models through Gemini CLI required developers to explicitly opt in by adding "experimental": { "gemma": true } to their settings.json file. The models were functional but gated behind the experimental namespace — a signal that the feature was not yet production-ready and could change without notice. With v0.42.0, that gate is removed. Gemma 4 access is now part of the standard configuration, and documentation references to the experimental status have been updated accordingly.

The Gemma 4 Model Family

The Gemma 4 family, launched in April 2026, includes four open-weight models released under the Apache 2.0 license. The lineup ranges from lightweight edge models — E2B (2B parameters) and E4B (4B parameters) — up to the flagship 26B MoE and 31B dense models, both available via the Gemini API. The 31B dense model placed third on Arena AI's global leaderboard at launch, outperforming models many times its size. All Gemma 4 models support 128K–256K context windows and native multimodal inputs including text, image, and video, with the smaller E2B and E4B variants additionally supporting audio.

What This Means for Developers

For Gemini CLI users on the Gemini API backend, Gemma 4 models are now accessible without any configuration change. Developers who previously tested Gemma 4 under the experimental flag will notice no behavioral change — but new users and those upgrading from older versions will find it available out of the box after running npm install -g @google/gemini-cli.

It is important to note that this change affects cloud-hosted Gemma 4 access via the Gemini API only. The gemini gemma setup command for local model routing — which downloads and runs a Gemma model on-device to make routing decisions — remains a separate feature that still requires explicit configuration.

Upgrade Path

Updating to v0.42.0 via npm install -g @google/gemini-cli is sufficient to receive the change. No settings.json modifications are required. Users who previously set "experimental": { "gemma": false } to suppress the feature should note that the explicit false value will continue to disable Gemma 4 access — they will need to remove or update that setting if they wish to enable it.