Woofun AI reports that Tencent has released compressed 1-bit and 4-bit versions of the Hy3 model, reducing its footprint from approximately 600GB to 85.5GiB and 169.9GiB respectively. The 1-bit variant enables single NVIDIA H20 deployment, while the 4-bit version requires two cards but offers performance closer to the original. Tencent notes that both versions maintain strong capabilities in code, tool invocation, and long-text understanding, with generation speeds increasing by roughly 50% and 60% when acceleration is enabled.