Ming-Omni
inclusionAI (Ant Group)Ming-Omni is inclusionAI's omni model line: it takes images, text, video and audio and answers in text, speech or generated images. Ming-flash-omni 2.0, the current release, is a mixture-of-experts model with 100B total and 6B active parameters on the Ling-2.0 architecture, and generates speech, sound and music in one pipeline with voice cloning, alongside image generation and editing. inclusionAI is Ant Group's AI lab.
The Ming-3.0 audio models, the Ming-omni-tts speech models and the Ming-Image design models share the name but are separate releases and are not part of this entry.
Openness
3 high confidence- weights
- open (Ming-flash-omni 2.0 on the Hub and ModelScope, ungated)
- data
- closed (no training data released for Ming-flash-omni 2.0)
- code
- partial (inference, vLLM and demo code, with no training scripts)
- license
- MIT (code and weights)
Ming-flash-omni 2.0's weights and the repository code are MIT. The repository runs the model but does not train it, and the training data is not released.
- https://huggingface.co/api/models?author=inclusionAI&limit=200&expand[]=downloads&expand[]=cardData&expand[]=gated&expand[]=createdAt recorded 2026-11-12
Ming-flash-omni-2.0 (created 2026-02-10), the newest Ming omni checkpoint, gated false and tagged mit.
- https://huggingface.co/inclusionAI/Ming-flash-omni-2.0/raw/main/README.md recorded 2026-11-12
Ming-flash-omni 2.0 card: license mit; input image, text, video and audio, output image, text and audio; no training data release.
- https://raw.githubusercontent.com/inclusionAI/Ming/main/LICENSE recorded 2026-11-12
LICENSE: MIT License, "Copyright (c) 2025 inclusionAI".
- https://ungh.cc/repos/inclusionAI/Ming/files/main recorded 2026-11-12
Repository tree: modeling, processing, test_infer.py, cookbook.ipynb and a vLLM patch; no training script or config.
Adoption
2 high confidenceHugging Face downloads over the trailing 30 days for the four declared omni checkpoints, most of them for Ming-flash-omni 2.0. The separately released Ming speech and audio models are not counted.
- https://huggingface.co/api/models?author=inclusionAI&limit=200&expand[]=downloads&expand[]=cardData&expand[]=gated&expand[]=createdAt recorded 2026-11-12
Ming-flash-omni-2.0 8,034, Ming-flash-omni-Preview 3,851, Ming-Lite-Omni-1.5 739 and Ming-Lite-Omni 215 downloads in the trailing 30 days (12,839 together).
Capability
5 high confidenceMing-flash-omni 2.0 understands audio and video natively alongside images and text and answers in speech, level with Qwen-Omni. It also generates and edits images, which does not change its place here.
- https://huggingface.co/inclusionAI/Ming-flash-omni-2.0/raw/main/README.md recorded 2026-11-12
Card: model table lists input "Image,text,video,audio" and output "Image,text,audio"; a streaming video conversation demo; unified speech, audio and music generation with zero-shot voice cloning.
- https://raw.githubusercontent.com/inclusionAI/Ming/main/README.md recorded 2026-11-12
README: Ming-flash-omni 2.0 released 2026.02.11, a mixture-of-experts model with 100B total and 6B active parameters on the Ling-2.0 architecture.
Verified 2026-11-12