Synced·Added· 14 hours ago
Hy4 flash is released: Inference cost reduced to 1/20, open-source performance tops
Brief overview
The Hy4 flash model reduces inference cost to 1/20 through Flash calls, achieves SOTA scores in multiple benchmarks, and completes development in 15 minutes. The model has a parameter scale of 175B, supports an 8k token context window, speeds up 3.5 times, and the cost per million tokens is only $3.5, it has been open-sourced and supports API calls.
Source: Synced · jiqizhixin.com