Gemini CLI: What's New in June 2026

This release primarily provides an update to the internal build version, linking to the full comparison log for detailed changes. Users should refer to the provided GitHub link for a complete breakdown of all modifications in this nightly update.

This release primarily consists of internal updates and maintenance, as indicated by the provided changelog link focusing on a specific nightly build comparison. Users should refer to the linked comparison for detailed commit-level changes between the two nightly versions.

This nightly release focuses primarily on a security enhancement. It enforces case-insensitive handling for sensitive path blocklists and improves the VS Code HITL mechanism.

This release focuses on stability and infrastructure updates. Key fixes include preventing bad NPM releases during CI and resolving a failing no_proxy test. Additionally, the Vertex AI base URL configuration has been updated.

This release introduces new capabilities, including support for GDC air-gapped Service Identity and the new eval:inventory CLI command with JSON output. Several critical bugs were fixed, addressing issues like zero-quota hang loops, path traversal vulnerabilities during skill install, and various CI/CD stability problems.

This release introduces the capability for tool registry discovery, enhancing how tools are managed. Several critical fixes were implemented to stabilize the release process, including preventing bad NPM releases and resolving binary shadowing issues during CI verification.

This release introduces significant new capabilities, including support for GDC air-gapped Service Identity and new CLI commands like "eval:inventory" with JSON output. Several critical bug fixes address path traversal vulnerabilities, resolve issues with pending tools and trust overrides, and stabilize the nightly release process.

This release introduces automatic selection of the 3.5 flash model when enabled via a flag, improving model routing efficiency. Key fixes include improved TOML parsing recovery and corrections to Vertex AI model mapping. Additionally, the display frequency of the Antigravity transition banner has been limited.

This release introduces support for GDC air-gapped Service Identity and includes a static evaluation source analyzer. Several critical bugs were addressed, including fixing infinite retry loops on zero-quota limits and preventing path traversal vulnerabilities during skill installation.

This release introduces new documentation and migration commands for the Antigravity CLI. Key fixes include ensuring zero-quota limits fail fast to prevent hangs and correcting multi-line escaped quote handling. Additionally, the tool now avoids persisting empty resume sessions and standardizes tool output formatting.

This release introduces the ability to include quotes in your interactions and enables the use of the flash GA model under specific experiment flags. Key fixes include hardening PTY resizing against native crashes and preventing spam loops caused by invalid editor configurations.

This release introduces automatic model selection, enabling the use of the 3.5 flash model when configured. Key fixes include improved TOML parsing recovery and corrected Vertex AI model mappings. Additionally, usability is enhanced by limiting the Antigravity banner display count and updating related documentation.

This release primarily consists of a patch update, cherry-picking a specific commit to stabilize the v0.46.0-preview.2 branch into the new v0.46.0-preview.3 version. No new user-facing features or specific bug fixes were detailed in this summary section.

This release appears to be a patch update (v0.45.3) created via a cherry-pick operation. No user-facing features or specific bug fixes are detailed in these brief notes.

This release focuses on minor housekeeping and user experience refinements. The primary change is limiting the display frequency of the Antigravity transition banner. Additionally, experimental text has been removed from the browser agent documentation.

This release appears to be a patch update (v0.45.2) based on cherry-picking a specific commit from the previous version branch. No user-facing features or specific bug fixes are detailed in this summary section.

This release primarily consists of a patch update (v0.46.0-preview.2) created by cherry-picking a specific commit (f40498d) from the previous preview version. There are no new user-facing features or specific bug fixes detailed in this summary section.

This release primarily consists of internal updates and minor refinements, as indicated by the focus on a comparison link without specific feature or fix descriptions in the main body. Users should refer to the provided link for detailed commit-level changes.

This release appears to be a patch update (v0.45.1) focused on cherry-picking a specific commit to stabilize the current version. No new user-facing features or specific bug fixes were detailed in the provided notes beyond the patch application.

This release introduces support for using the Gemini 1.5 Flash model automatically when enabled via a flag. It also includes stability improvements by adding recovery mechanisms for policy parsing errors. Additionally, the continuous integration process has been optimized with new labeling and batching workflows.

This release primarily consists of a patch update, cherry-picking a specific commit (e4315b3) to stabilize the v0.46.0-preview.0 branch. Users should note this is a maintenance release focused on applying necessary fixes to the current preview version.

This release introduces the exposure of usage metadata and completes significant simplification work in context handling. Several critical bugs were addressed, including fixing Termux relaunch loops, ensuring sequential execution of the update_topic tool, and resolving orphaned function response errors caused by routing classifiers.

This release introduces the ability to transition to the flash GA model when an experiment flag is active. Key fixes include hardening PTY resizing to prevent native crashes and resolving a spam loop caused by an invalid preferredEditor setting. The update also adds support for including quotes in the output.

This release introduces a significant change by enabling the transition to the flash GA model when the appropriate experiment flag is set. This update focuses on leveraging the new model architecture for potentially improved performance or capabilities.