Added Claude Sonnet 5.5 (Anthropic).

Raised the chat token budget from 8,192 to 32,000 for all cloud providers. Recent models spend part of their output budget thinking or reasoning before they answer, which left little room for the reply itself.

No new permissions and no new network hosts.

Three new AI models:

• Claude Opus 5.5 (Anthropic)
• GPT-6 Sol (OpenAI)
• GPT-6 Luna (OpenAI)

No other changes. No new permissions.

Added tooltips to the header buttons, and a link in Settings to the API key page for whichever provider you have selected.

The empty output pane now explains how to start a build rather than only where files will appear.

Fixed: typing a bracket or parenthesis in the model search broke the filter. The Terms of Service link pointed at a page that has moved. The help guide named buttons in English, or in words the interface does not use, in the eight non-English languages.

Added GPT-6 Astra to the model list.

Added Gemini 3.8 Flash to the model list.

Added Claude Fable 5.1 to the model list.

Added Gemini 3.7 Flash and Claude Fable 5.

Fixed a bug where an API key entered while a custom or on-device model was selected could be saved to, or deleted from, the wrong provider's settings. If you have keys saved for more than one provider, it is worth checking that each is stored under the right one.

Google requests now send your API key in a request header rather than in the request URL, so it is no longer exposed in server and proxy logs.

Corrected the in-app help: model selection now comes before entering an API key, which is the order the extension actually requires, and the help now notes that on-device models need no key and cannot build extensions.

Added seven new AI models: Gemini 3.6 Flash and Gemini 3.5 Flash Lite (Google), GPT-5.6 Sol, GPT-5.6 Terra and GPT-5.6 Luna (OpenAI), and Claude Opus 5 and Claude Sonnet 5 (Anthropic).

Removed four models that are no longer usable: GPT-5.4 Pro and GPT-5.5 Pro, which are not available on the chat completions endpoint, and Gemini 2.5 Flash and Gemini 2.5 Pro, which Google no longer serves to new users.

Fixed OpenAI requests so the GPT-5 family works again — those models require a newer token parameter than the extension was sending.

Fixed a bug where the model shown in the selector could differ from the model actually used, which could cause an API key to be saved under the wrong provider.

Added three new AI models: Gemini 3.5 Flash, GPT-5.4 Pro, and Claude Opus 4.8.
Increased the maximum output length for Claude Haiku 4.5 so longer generations are no longer cut off early.
Removed GPT-5.3 Instant, which the provider is retiring.

Version 2.1.0

New features:
- Import from ZIP: Load existing browser extensions or websites from a ZIP file to view and edit with AI

Improvements:
- On-device AI models are now correctly hidden on Firefox
- Updated Terms of Service version check
- All 9 language translations updated

No breaking changes. All existing features work as before.