← All models

VTP-Small-f16d64

Published Publisher MiniMax

Overview

VTP-Small-f16d64 is the Small checkpoint in the VTP visual-tokenizer research project involving MiniMax and Huazhong University of Science and Technology. It is intended for image feature extraction and visual representation learning, rather than conversational text generation. The model card lists approximately 0.2 billion parameters. (huggingface.co)

Capabilities and uses

VTP combines contrastive, self-supervised and reconstruction learning. The project demonstrates image encoding and reconstruction, CLIP-style image–text comparisons, and extraction of visual features for classification. Visual tokens can also serve as inputs to a separately trained diffusion model for image generation. The repository’s illustrative inference and generation examples use the Large checkpoint, so their reported results should not be attributed to the Small checkpoint. (github.com)

Limitations and access

The project describes itself as work in progress. Image generation requires a separate generation pipeline; the checkpoint alone is not presented as a standalone text-to-image service. The model card labels the license modified-mit; review its terms before use. The checkpoint is available through the official model repository. (huggingface.co)

Release date

December 16, 2025. The official project announced its technical report and pretrained weights on that date. (github.com)

Pricing

As of November 17, 2026, pricing for this specific checkpoint, including input and output billing, was not verified on either MiniMax’s mainland China or international official pricing page. No CNY or USD amount is assumed, and prices for other models do not apply to this checkpoint. (platform.minimax.cn)

Official sources