Seamlessly integrate and switch between language and vision models using the mwai_ai_query filter. The primary goal of this guide is to demonstrate the implementation of a dynamic model-switching mechanism, allowing you to send the heavier model only when it is actually needed. Note that modern models handle images natively, so you no longer need a separate vision model; this example now switches to a faster and cheaper model when the request is a simple one. By incorporating this approach into your code, you can optimize resource usage and enhance the efficiency of your applications that leverage AI capabilities.

add_filter( 'mwai_ai_query', function ( $query ) {
		// If it's an embbeding request, return the query as is
	  if ( $query->mode == "embedding" ) {
	    return $query;
	  }

    // Check ithe type of uploaded file
    $needVision = ( $query->filePurpose === 'vision' );

    // Switch the model based on the need for vision capabilities
    if ( $needVision ) {
        $query->set_model( 'gpt-5.5' );
    } else {
        $query->set_model( 'gpt-5.4-mini' );
    }

    // Log the chosen model to error log for debugging purposes
    error_log("🗒️: " . print_r( $query->model, true ) );
  
    return $query;
}, 10, 2 );