According to ChainCatcher, Tencent Hunyuan has launched Hy4 Flash, its first model with “Flash” in the name, adding a lower-cost tier alongside flagship Hy4. Hy4 Flash has 224 billion total parameters, including 64 billion Engram memory parameters, and activates 12 billion per inference, compared with Hy4’s 770 billion total and 49 billion active parameters. It supports multimodal input and a 1-million-token context window.
Engram, a conditional memory technology previously proposed by DeepSeek, stores common text patterns in searchable memory tables to reduce repeated computation. Hy4 Flash is priced at 0.5 yuan per million input tokens, 2 yuan per million output tokens, and 0.02 yuan for cached tokens—12, nine, and 15 times lower, respectively, than Hy4’s list prices. Hy4 currently has a limited-time 50% discount, making the actual price differences smaller.
In Tencent’s tests, Hy4 Flash scored 74.0 on collaborative PPT editing, down from Hy4’s 78.9, while task cost fell from 1.714 yuan to 0.105 yuan. On automated tasks, its score was 52.2 versus 56.2, with cost per task dropping from 0.908 yuan to 0.038 yuan. Hy4 Flash scored 19.2 on Terminal-Bench 4.0, compared with 44.4 for Hy4, and 16.0 on ProgramBench, versus 30.0 for Hy4.