* Add: Support for GPT-6 Astra, including its max reasoning level.
* Fix: Temperature is no longer sent to models that reject it.
* Update: The reasoning dropdown now only shows levels supported by the selected model.
* Add: Claude Fable 5.1 to the Anthropic models.
* Fix: Anthropic now caps max_tokens to the model's limit instead of returning a 400 error.
* Fix: Provider errors with a flat body (such as Mistral) and any 4xx/5xx status now show the actual error message.
* Update: Removed the Assistants module following OpenAI's shutdown.
* Update: Transcription now uses gpt-transcribe, and Opus 4.1 has been removed.
* Update: Built-in tools that can't be used with Chat Completions are now reported in the logs instead of being dropped silently.
* Fix: Sending a message after a client-side JS function call no longer fails with an invalid_request_error.
* Fix: MCP OAuth login redirects are no longer cached by page caches and served to the wrong visitor.
* Fix: Plugin file tools no longer report a refusal as a missing directory.
* Update: The "Done!" placeholder for client-side function calls is now translatable and filterable.
* Fix: Chatbots using the default environment and model are no longer excluded from the function calling list.
* Fix: Vision now detects image types from the file contents instead of the extension, so images served through a CDN work.
* Fix: Magic Wand can read the selected text again in the iframed block editor of WordPress 7.1, re-enabling Suggest Synonyms.
* Update: Completed steps in the Setup Assistant now collapse to a single line.
* Fix: Editor Assistant endpoints accepted a nonce any visitor could generate, allowing anyone to run AI queries on the site's provider account (reported by Abdullah Kareem via WPScan).
* Fix: A crafted URL could escape the uploads folder and send any server-readable file to the AI provider (reported by Jashid Sany via WPScan).
* Fix: Editor Assistant now works for admin-equivalent accounts whose role isn't literally administrator.
* Fix: Embeddings sharing the same reference no longer stay stuck, and a single unprocessable vector no longer blocks the maintenance queue.
* Fix: A temperature of 0 is no longer treated as unset and stripped from the request.
* Fix: Image-only replies from GPT-5 models no longer show "medium" as the answer text.
* Update: Claude Sonnet 5 pricing ($2/$10 is now permanent), dashboard wording, and tested with WordPress 7.1.
* 🎵 Discuss with others about Ai Engine on [the Discord](https://discord.gg/bHDGh38).
* 🌴 Keep us motivated with [a little review here](https://wordpress.org/support/plugin/ai-engine/reviews/). Thank you!
* 🥰 If you want to help us, check our [Patreon](https://www.patreon.com/meowapps). Thank you!
* 🚀 [Click here](https://trello.com/b/8U9SdiMy/ai-engine-feature-requests) to vote for the features you want the most.
* Add: WPML support in the MCP server, mirroring the Polylang tools.
* Add: SEO section in the Modules tab, with robots.txt access for AI crawlers and live stats when SEO Engine is installed.
* Update: MCP self-test now detects AI-crawler blocking, nested .htaccess files and www mismatches, and logs every OAuth authorize refusal.
* Update: Internal errors are now hidden behind a filterable public message, and AI Forms no longer show provider errors to visitors.
* Update: Shared dashboard synced with the new Board and the AI site analysis.
* Fix: The mwai_mcp_mutate hook now fires on deletions, so cache purges hooked to it run.
* Fix: Qdrant collections were always created with 1536 dimensions regardless of the embedding model, and default dimensions are now read from the right place.
* Update: Embedding dimensions are no longer locked for OpenAI-compatible models, as that size is only a guess.
* Fix: The MCP access level no longer restricts OAuth connections, and the setting is no longer hidden behind the bearer token.
* Fix: MCP connections no longer drop at random when the client refreshes its access token.
* Fix: Workspace mobile app connection on servers that don't expose the Authorization header, with clearer pairing errors.
* Fix: A Context Max Length of zero no longer empties posts, which blocked embeddings sync and blanked chatbot context.
* Update: Transcription now defaults to gpt-4o-mini-transcribe instead of whisper-1.
* Update: The mobile connection warning links to the troubleshooting page instead of showing the htaccess snippet.
* Add: Logging on the MCP OAuth token endpoint for failing refreshes.
* Add: Workspace is officially launched on [iOS](https://apps.apple.com/app/workspace-for-wordpress/id6794717714).
* Update: OpenAI pricing for GPT-5.6 Terra and Luna following the July 30 cut, and corrected o3 which was priced 7.5x too high.
* Update: Removed the leftover mcp.js relay and its documentation, and corrected the MCP header comment that still mentioned SSE and OAuth.
* Add: getDiscussion() to the PHP API, to read a stored discussion without using $mwai_core.
* Update: Minimum PHP version is now 8.1, declared in the plugin header so WordPress blocks activation on older versions.
* Fix: Push All no longer exhausts the memory limit on large sites; post content is now checked in chunks.
* Fix: Long Japanese and Chinese posts no longer produce empty content, which left their embeddings stale.
* Fix: Hardened the REST endpoints against client-supplied file paths and API keys, corrected the MCP OAuth route match and guest session cookie, and stopped the chatbot key reaching Editors.
* Fix: The mwai_mcp_callback filter now receives five arguments from both the Workspace and the MCP endpoint.
* Add: CSV and JSON export for the Query and MCP logs, honouring the active filters and sort.
* Fix: Chatbot input is no longer hidden behind the mobile keyboard in fullscreen, and shortcuts are now chips that don't inherit the site's button styles.