Trident-Qwen3VL-2B: an open multimodal embedding model for mixed-modality retrieval

OpenBMB releases Trident-Qwen3VL-2B, a multimodal embedding model that maps text, images, and fused text-image inputs into one shared space for mixed-modality retrieval. It is fine-tuned from Qwen3-VL-2B-Instruct on Trident-Train, with the LoRA adapter already merged. The Trident-specific weights are released under the MIT License, while the base model's Apache-2.0 terms still apply.