Model Positioning
Tencent Hunyuan's video understanding model based on turbos-vision; Tencent Cloud TokenHub lists this version as HY-Vision-Video. (cloud.tencent.com)
Capabilities and Uses
Supports video description and question answering based on video content, and can be used for scenarios such as video content analysis and question answering. The TokenHub example uses a video URL and a text question as input and generates a text answer. (cloud.tencent.com)
Limitations
- TokenHub lists a context window of 32k tokens, a maximum input of 24k tokens, and a maximum output of 8k tokens. (cloud.tencent.com)
- The TokenHub documentation states that this model does not support video input via Base64; a video URL directly accessible on the public internet must be used. Refer to the API documentation for other restrictions on video format, duration, and size. (cloud.tencent.com)
- The model is used to understand video and generate text; it should not be confused with a video generation model.
Release Date
July 28, 2025. (cloud.tencent.com)
Pricing
Tencent Cloud mainland China TokenHub online inference list price: input 3 Chinese yuan (CNY) per million tokens, output 9 Chinese yuan (CNY) per million tokens. This price applies to HY-Vision-Video as listed on the platform; actual charges are subject to the platform's billing rules and bills. Date checked: November 18, 2026. (cloud.tencent.com)
Official Sources
- Tencent Cloud product update announcement: The original link timed out; the content was read from the corresponding official Tencent Cloud mirror page. ()
- TokenHub model list: Model ID, capabilities, and context limits. (cloud.tencent.com)
- TokenHub video understanding invocation guide: Video input methods and limitations. (cloud.tencent.com)
- TokenHub mainland China model pricing: Date checked: November 18, 2026. (cloud.tencent.com)