The ALoDLM 8b looped diffusion language model runs roughly three times faster than its base Qwen3-8b model and achieves better benchmark results. The claim was shared on Reddit’s r/LocalLLaMA by user billy_booboo on October 16, 2026.