OmniEmbed v0.1
Tevatron/OmniEmbed-v0.1
published Apr 2025 · updated Apr 2026
OmniEmbed v0.1 is a multi-modal embedding model that unifies text, image, video, and audio retrieval, built on Qwen2.5-Omni-7B.
specs
| Task | Visual-Document-Retrieval (Multi-modal Retrieval) |
| Architecture | Qwen2.5-Omni-7B |
| Parameters | 7B |
| License | Apache-2.0 |
about this model
OmniEmbed is a multimodal embedding model for visual-document-retrieval, built on Qwen2.5-Omni-7B via the Tevatron toolkit. It generates unified embeddings across multilingual text, images, audio, and video, enabling cross-modal retrieval. The model is described as the first embedding model to unify all four modalities and has been accepted at SIGIR 2025 (Demo track).
OmniEmbed achieves strong results on standard benchmarks, performing comparably to models optimized for individual tasks:
| Benchmark | Task | Metric | OmniEmbed | Baseline |
|---|---|---|---|---|
| BEIR-13 | Text Retrieval | nDCG@10 | 58.2 | MistralE5 (59.0) |
| MIRACL | Multilingual Retrieval | nDCG@10 | 69.1 | BGE‑M3 (69.2) |
| VIDORE | Image Document Retrieval | nDCG@5 | 85.8 | DSE‑QWen2 (85.8) |
| MSRVTT | Video Retrieval | R@1 | 51.3 | CLIP (31.2) |
| AudioCaps | Audio Retrieval | R@1 | 34.0 | 23.1 |
The underlying model is a LoRA fine-tuned version of Qwen2.5-Omni-7B. Training data and code are fully open-source. As a hosted API on gigarouter, OmniEmbed requires no local installation—simply call the OpenAI-compatible endpoint to embed queries and documents of any supported modality.
best for
- ·Cross-modal search across text, images, audio, and video
- ·Multilingual text retrieval
- ·Image document retrieval (charts, PDFs, screenshots)
- ·Video retrieval for tutorials or instructional content
FAQ
It supports text, image, audio, video, and unified multimodal inputs (e.g., text+video).
It achieves 51.3 R@1 on MSRVTT, significantly outperforming CLIP (31.2).
It is built on Qwen2.5-Omni-7B with 7 billion parameters.
The Tevatron toolkit is Apache-2.0; OmniEmbed follows the same license.
Use the gigarouter OpenAI-compatible endpoint with your API key, following standard embedding API conventions.
We're benchmarking and onboarding OmniEmbed v0.1 as a hosted, OpenAI-compatible API. Sign in for free credit and be ready when it lands, or tell us you want it and we'll prioritize it.