siglip-so400m-patch14-384

Architecture

Shape-optimized ViT (SoViT-400M/14) image encoder at 384px with a paired text encoder, trained with the sigmoid (SigLIP) loss instead of softmax contrastive — which scales better and improves zero-shot accuracy at a given compute budget.

Key Capabilities

• •High-accuracy image+text embeddings (sigmoid loss)
• •Strong fine-grained visual retrieval and zero-shot classification
• •Efficient 400M shape-optimized backbone
• •384px input for detailed scenes and on-image text

Use Cases on Mixpeek

• •Fine-grained product and scene visual search
• •Zero-shot tagging where CLIP ViT-L is not accurate enough
• •Video keyframe embeddings for media retrieval
• •Cross-modal recall feeding a reranker

Use siglip-so400m-patch14-384 on Mixpeek

Build multimodal processing pipelines with this model and others. Extract features, run inference, and set up retrieval in Mixpeek Studio.

On Mixpeek, siglip-so400m-patch14-384 runs as a managed extractor inside a processing pipeline. Point a bucket of zero shot image classification data at it, and Mixpeek handles GPU provisioning, batching, retries, and writing the outputs into a vector store you can query.

Extractor outputs land in the Mixpeek Vector Store (MVS), where you can combine them with retrieval, reranking, and filter stages to build end-to-end search and agent-perception pipelines, no model-serving infrastructure to maintain.