CLIP vs YOLOv8 Classification

OpenAI CLIP (Contrastive Language-Image Pretraining) is a vision-language model released in January 2021 by OpenAI. It jointly trains an image encoder and a text encoder to produce matching embeddings for image-caption pairs, using a contrastive objective over WebImageText (WIT), a dataset of 400 million image-text pairs collected from the public web. By learning to associate images with free-form text rather than a fixed set of class labels, CLIP produces a shared embedding space that enables zero-shot classification with arbitrary vocabularies at inference time.

CLIP supports zero-shot image classification by embedding candidate class labels as text and selecting the label whose embedding is closest to a given image's embedding. It is also widely used for image-text retrieval, as a frozen backbone in downstream vision-language models, and as a building block for content moderation, similarity search, and generative model guidance — notably as the text conditioning mechanism in early versions of Stable Diffusion. OpenAI released several CLIP variants built on different vision encoders, including ResNet and Vision Transformer backbones at multiple sizes and input resolutions, with ViT-L/14 at 336 pixels being the largest and most widely adopted. CLIP is distributed under the MIT license. The model has been widely influential as the basis for subsequent vision-language work — including SigLIP, OpenCLIP, and MetaCLIP — and remains a common reference baseline despite being released in 2021 and surpassed on many benchmarks by later models.

YOLOv8 Classification is the image classification variant of the YOLOv8 model family from Ultralytics, released in January 2023. Unlike the primary YOLOv8 detection and segmentation models, which predict bounding boxes or pixel masks, YOLOv8 Classification predicts a single class label for a full input image, supporting standard single-label image classification tasks. It shares the YOLOv8 backbone architecture, including the C2f (Cross-Stage Partial with 2 convolutions) module, with the detection variants, making it straightforward to use within the same Ultralytics training and inference workflow as other YOLOv8 tasks.

YOLOv8 Classification is released at five sizes: YOLOv8n-cls (nano, 2.7M parameters), YOLOv8s-cls (small, 6.4M), YOLOv8m-cls (medium, 17.0M), YOLOv8l-cls (large, 37.5M), and YOLOv8x-cls (extra-large, 57.4M). These variants allow users to trade off accuracy against inference speed and memory footprint. Pretrained checkpoints are provided for ImageNet classification at 224 pixel resolution, and the model can be fine-tuned on custom datasets using the Ultralytics Python API or command-line tools. The model supports export to common deployment formats including ONNX, TensorRT, CoreML, and TensorFlow Lite. YOLOv8 Classification is distributed under the AGPL-3.0 license, with an Enterprise License available from Ultralytics for proprietary deployments. The YOLOv8 family has since been succeeded by YOLO11 (September 2024) and YOLO26 (January 2026), each of which includes equivalent classification variants.