Moore Threads has completed full‑stack inference support for the open‑source
biomolecular structure model Protenix‑v2 on its MTT S5000 inference card using
the in‑house MUSA software stack. Benchmarks show the S5000 delivers roughly 26%
average end‑to‑end inference speedup versus mainstream international GPUs, with
3D conformation prediction accuracy closely aligned. Protenix‑v2, developed by
ByteDance’s Seed team and released in April 2026, is an open‑source foundation
model capable of high‑accuracy end‑to‑end prediction for protein, DNA, RNA and
small‑molecule ligand complexes and removes limits on non‑open overseas model
weights and private deployment.