Dark Mode for Visual Picker: The side panel of the Visual Rules picker now fully respects your system's dark mode settings. When browsing on a dark-themed OS or browser, the panel will automatically switch to a matching dark surface, preventing stark contrast issues during nighttime browsing sessions.

Fixed Site Suggestions: A critical bug in release builds has been resolved where the "Add current site" suggestions in the blocklist and Visual Rules picker would silently fail to detect the open webpage. The input fields will now reliably display the correct current-tab hints across all builds.

Visual Rules Picker: A new interactive picker has been added to Settings → Rules, allowing you to visually click elements on any webpage to define what should be translated or skipped. The tool automatically handles domain formatting to ensure accurate scoping, while a raw JSON editor remains available for power users.

Redesigned Rules Management: The blocklist and custom site rules have been completely overhauled into clean, manageable lists instead of monolithic text blocks or single JSON blobs. You can now easily filter through entries with a typo-tolerant search, pause individual rules or hostnames without deleting them, and catch formatting errors instantly when adding new patterns.

Unified Rules Dashboard: All translation controls—including the blocklist, custom site rules, and translation topics (glossaries and instructions)—are now consolidated into a single, intuitive section. Deep links and bookmarks from previous versions will automatically redirect to this new unified dashboard.

Smarter Custom Rule Editing: Editing an existing site rule now safely opens the visual picker in a new tab to preserve your current workflow, preloading your saved selectors for quick adjustments. Rules that apply to multiple sites are neatly organized into a dedicated multi-site fold to keep the main list clutter-free.

Polished Interface Layout: The backup configuration import/export options have been moved to a more logical position under the Extras panel next to cache controls. Additionally, the initial onboarding screen has been polished with a flat background to ensure perfect visual consistency with the rest of the extension's design.

Native Anthropic Claude Support: Anthropic Claude is now a built-in engine option listed right after OpenAI, allowing you to paste your API key and translate immediately without any custom profile setup. The picker defaults to Claude Haiku 4.5 for fast and cost-effective translations, and it automatically fetches the live model lineup directly from the Anthropic API.
Gemini 3.5 Flash Integration: Gemini 3.5 Flash is now selectable in the Gemini model picker following its official release. While Flash-Lite remains the default due to token cost efficiency, you can easily switch to this stronger new variant from the model dropdown menu.
Tailored Subtitle Translation: LLM-powered translation now utilizes a dedicated subtitle prompt to keep text structure natural and prevent stiff punctuation in auto-generated YouTube captions. The new prompt perfectly preserves speaker labels or bracketed stage directions, while a smart secondary variant seamlessly corrects speech-recognition errors.
Dynamic YouTube Styling: The translated subtitle overlay now automatically synchronizes with YouTube’s native caption font size and family settings. This allows the translated text to scale beautifully alongside the original captions instead of competing for attention at a fixed size.
Clearer Onboarding Flow: The initial onboarding step has been updated for all users to explicitly mention that advanced engines like DeepL, OpenAI, and Gemini can be configured later in Settings. This ensures new users are fully aware of the wide range of available translation services from the start.
Firefox Platform Cleanups: The On-Device AI engine and its related feature toggles have been cleanly removed on Firefox since the browser does not expose the required built-in AI APIs. If imported from a Chrome configuration, the extension will automatically switch to Google Translate to keep everything working smoothly.
Smarter Error & Config Handling: The browser popup now visually flags and disables unconfigured engines with a "Not configured" tag to prevent translation failures. If an unconfigured engine is triggered, a localized error toast will now clearly name the specific service and point you directly to the Settings page to fix it.
Streamlined Usage Panel: The Usage section now displays a clean sign-in placeholder instead of empty metrics or fabricated page allowances when the active engine is Cleartext but you are logged out. Clicking the placeholder provides a direct, one-click jump straight to the sign-in form.

Free Microsoft Translator: You can now use the Microsoft Translator engine completely free without an Azure API key, leveraging the same built-in endpoint as Microsoft Edge. If you prefer your paid Azure subscription, you can still paste your key to use your dedicated endpoint as before.

Smart Language Skipping: The translation engine now intelligently skips short English snippets on pages you've already chosen to ignore, such as Google Search AI Overviews. Additionally, the PDF viewer will now pre-detect languages per paragraph, preventing wasted API calls on text that already matches your target language.

Broader Content Translation: Web components and content tucked inside open Shadow DOMs, such as custom CMS embeds and data visualizations, are now fully detected and translated alongside the rest of the page.

Improved PDF Viewer: The PDF viewer's error screens have been completely redesigned into localized, user-friendly status pages. Please note that on Firefox, PDF translation is now focused exclusively on online web PDFs, as local file:// PDFs will remain in Firefox's built-in viewer due to platform limitations.

Searchable Model Pickers: The OpenAI and Gemini model dropdowns now permanently display a search box to help you find specific model variants faster. The Cleartext engine's model preference dropdown has also been upgraded to a styled picker that elegantly highlights Max-only models.

Seamless Auto-Translate & Fixes: Auto-translate preferences now correctly match websites regardless of whether the www. prefix is present, and these site lists can now be exported or imported. Furthermore, a text-rendering bug was fixed to ensure Google Translate output no longer displays raw HTML entities like apostrophes or quotes.

Smart Translation Toggle: The page translation shortcut now acts as a two-way toggle, allowing you to easily switch back and forth between the translated version and the original text. The default shortcut has also moved to Alt+A (⌥A on macOS) to prevent system conflicts and ensure reliable binding.

Streamlined Settings: All preference changes and API key edits are now saved automatically in the background within a third of a second. Additionally, the translation engines are now grouped into Recommended, Translation Services, and AI Translators for much easier navigation.

Carbon Color Palette: The interface has shifted to a neutral-gray theme that removes any yellow undertones. Text contrast has been slightly softened to provide a more comfortable and less tiring reading experience during long sessions.

Enhanced PDF Support on Firefox: Full PDF translation is now supported on Firefox for both online links and local file:// documents. The extension now intercepts these documents early to open them seamlessly within the native cleartext viewer.

Unified Custom AI Engine: The custom engine setup is now consolidated into a single profile that automatically detects OpenAI and Anthropic protocols. It smartly identifies and routes individual models on mixed gateways, while updating your settings status dynamically.

Polished Interface Details: The custom model selector now leads with a searchable dropdown menu to reduce clutter and make model selection effortless. Minor visual bugs, such as height mismatches and background page-bouncing on macOS, have also been fixed.