Major labs continue to advance autoregressive transformer models like those from Anthropic, OpenAI, and Google, which dominate Chatbot Arena and other leaderboards through superior scaling, training data, and capabilities on complex reasoning and knowledge tasks. Discrete diffusion LLMs (dLLMs) such as LLaDA, Dream, and recent variants like d3LLM or Fast-dLLM show promise for parallel generation and speedups of 2-5x on certain benchmarks, yet remain limited to smaller scales (under 26B parameters) with performance gaps in overall quality and long-context handling. With only months left before the end of 2026, no dLLM has approached frontier status. A breakthrough scaling effort or hybrid approach from a major lab could still shift outcomes, though technical hurdles in training stability and inference consistency make this improbable.