Background image processing and refreshed AI models headline ClassifAI 3.9.0

ClassifAI 3.9.0 adds a new background processing mode for images, and refreshes the default models used across OpenAI, Google, and xAI. It also fixes a bug that could leave the settings screen showing a blank page, and bumps the minimum required WordPress version from 6.8 to 6.9.

What’s new for site owners

The headline addition is a new background processing mode for image features. Previously you could either run image analysis automatically (which blocked the upload request until it finished) or trigger it manually per image. The new mode runs automatically in the background using Action Scheduler, so uploads stay snappy while ClassifAI works through the queue behind the scenes.

The excerpt generator has also been updated to integrate directly with WordPress’s modern excerpt panel, rather than replacing it with a custom one. You’ll get a more native editing experience that matches how the rest of the block editor now behaves.

• Refreshed default models — OpenAI features now default to GPT-5.4 mini and gpt-image-2, Google image generation uses Nano Banana 2, and xAI features use Grok 4.3.
• Rewritten default prompts — All built-in prompts have been made more robust to take advantage of the much larger context windows in modern models.
• Better previewer feedback — Terms in the classification previewer that fall below your configured threshold are now clearly styled and accompanied by an explanatory notice.
• More reliable Text to Speech — The Classic Editor integration now polls for completion so you no longer need to reload the page, and existing audio files are preserved when content hasn’t changed.

Bug fixes and stability

Several fixes in this release are worth calling out. The settings page no longer crashes to a blank screen when a particular JavaScript object isn’t defined, user role handling has been hardened to prevent a potential fatal error, and default prompts now render with proper line breaks in the settings UI. The Title Generation modal and admin styling for Key Takeaways and Text to Speech have also been tidied up.

For developers

There are several changes developers should be aware of:

• Google image request body changed — Because the default Google image model moved to Nano Banana 2, the payload passed through the classifai_googleai_images_request_body filter has changed; update any code hooked into that filter.
• New get_prompt method — All built-in prompts have been moved into their own files and are now retrieved via a new get_prompt method, making it easier to override or extend them.
• AWS SDK prefixing — The bundled AWS SDK is now dependency-prefixed to prevent conflicts with other plugins that ship the same package.
• OpenAI API — Requests now use max_completion_tokens instead of the deprecated max_tokens parameter, and Key Takeaways plus Image Tags Generator now use the response_format parameter for more reliable structured output.
• Requirements — Minimum WordPress version is now 6.9, and the plugin is tested up to WordPress 7.0.
• Tooling — E2E tests have migrated from Cypress to Playwright, PHPStan coverage has been raised to level 5, and the codebase now aligns with the WordPress Coding Standards directly.

See the full release notes on GitHub. Learn more and download ClassifAI.