Tencent has announced the open-sourcing of its Hunyuan Hy4 Flash model, which features 220B total parameters but activates only 12B per token—a sparse activation ratio that serves as its core selling point. The Hunyuan team bills it as the most cost-effective Hy model to date, with the average cost per task coming in at just 1/20 that of Hy4 and 1/8 that of DeepSeek-V4.1-Flash-1017. Compared with the previous-generation Hy4, the model incorporates three structural changes.
