Tencent launches Hy4 flash at half price through November 30
Edited and reviewed by AI Ping Editorial Team
3-line summary
- Tencent added its lower-priced Hy4 flash model to TokenHub.
- International list prices are $0.07 per million input tokens and $0.281 per million output tokens.
- The half-price offer ends at noon China Standard Time on November 30.
Tencent added Hy4 flash, a lower-priced model in its Hunyuan family, to its cloud API on November 16. Tencent Cloud’s international TokenHub lists the model at $0.07 per million input tokens and $0.281 per million output tokens. Tencent is charging half those rates for usage through noon on November 30 in China Standard Time, or 1 p.m. in South Korea.
The larger Hy4 model, which arrived on TokenHub on November 4, remains available. Developers can send frequent requests such as document classification or chart interpretation to the lower-priced model and keep Hy4 for more involved jobs through the same service. TokenHub calls the new model `hy4-flash`; the existing model is `hy4`.
The launch price changes on November 30
TokenHub bills for input and output separately. Hy4’s international list rates are $0.834 per million input tokens and $2.501 per million output tokens. At list price, Hy4 flash costs about one-twelfth as much for input and one-ninth as much for output. The cost of each request still depends on how many tokens the model receives and produces.
Tencent applies the offer to usage incurred during the discount period and returns the model to list pricing when it ends. Hy4 has a separate half-price offer through the end of December 31 in China Standard Time. That leaves the two models on different billing schedules in December, even though both are offered at a discount in November.
Input served from a cache has its own rate. Tencent’s international table lists cache hits for the Flash model at $0.003 per million input tokens. In the Chinese price table, the corresponding online inference rates are 0.5 yuan for input, 2 yuan for output and 0.02 yuan for cache hits per million tokens. For an application that sends the same instructions or report template repeatedly, the share of input that hits the cache can affect the total cost.
The model accepts images alongside text
Tencent’s API guide gives both Hy4 flash and Hy4 a context window of one million tokens. It lists a maximum input length of 960k tokens and a maximum output length of 64k tokens. A service that forwards a long document along with an existing conversation has to fit the material within the request’s input limit.
The Flash model can read images mixed with text and respond in writing. An application could send a chart from a monthly report with a question about which categories fell. Tencent supports the request formats used by OpenAI Chat Completions, OpenAI Responses and Anthropic Messages. Tencent’s service handles the inference and billing for its API.
The guide sets a 10 MB limit per image and a maximum of 600 images per request, including images carried over from earlier turns. Image detail also affects the amount of input the model processes. A low-detail setting scales an image to fit within 512 by 512 pixels. The higher-detail option is aimed at material such as small print and tables.
Tencent positions the smaller model for routine office work and frequent calls. Its catalog lists support for calling functions and returning structured responses as well as recognizing images. An application can use those features to send a chart reading to another part of a workflow. Tencent’s API documentation still describes Hy4 as the model aimed at longer tasks and sequences of actions.
The coding tool OpenCode also lists a limited-time free entry under `hy4-flash-free` in its own Console catalog. OpenCode says the provider serving this entry does not retain request data or use it for training. The free offer applies to calls through OpenCode’s service, which gives no specific end time on its model page.
Sources
- TokenHub product updates — Tencent Cloud
- Model pricing (international) — Tencent Cloud
- TokenHub model pricing (China) — Tencent Cloud
- Hy API Guide — Tencent Cloud
- TokenHub model list — Tencent Cloud
- TokenHub product page — Tencent Cloud
- Console models — OpenCode
- Image Understanding — Tencent Cloud
