Resemble AI has launched DETECT-3B Omni, a multi-modal deepfake detection model that addresses the growing threat of synthetic media across audio, images, and video. This model integrates advanced audio detection capabilities from its predecessor, DETECT-2B, with a new vision component trained on millions of images and videos. The result is a robust 3 billion parameter model that boasts state-of-the-art detection accuracy: below 6% equal error rate (EER) for audio, around 9% for images, and approximately 4.5% for video. The model can identify deepfakes across over 40 languages and is equipped to handle various audio codecs, ensuring reliable detection in real-world conditions. The audio component has seen significant enhancements, including a broader training dataset that covers diverse noise conditions and improved performance on telephony codecs like G.711. The image detection features high accuracy against outputs from popular generative models such as StyleGAN 3 and DALL-E 3, detecting manipulated images with over 85% accuracy. Video performance is particularly strong against new generation models, with detection rates exceeding 99% on Veo 2. DETECT-3B Omni is accessible via the same API as its predecessor, allowing for straightforward integration. Users can submit audio, image, and video files for analysis, with the system automatically routing content to the appropriate detection model. For organizations with specific data residency needs, on-premise deployment options are available. Future developments will focus on expanding video coverage, real-time detection capabilities, and enhancing defenses against adversarial attacks targeting the detection system.