Orthrus-Qwen3 introduces a dual-architecture model boosting inference speed up to 7.8× while maintaining exact output fidelity, using a novel diffusion-based approach.
Orthrus-Qwen3: up to 7.8×tokens/forward on Qwen3, identical output distribution
