DreamReasoner-8B: Block-Size Curriculum Learning for Diffusion Reasoning Models

DreamReasoner-8B is an open-source block diffusion reasoning model that employs a block-size curriculum learning approach to enhance long chain-of-thought reasoning. The model demonstrates that training with small block sizes significantly improves reasoning performance compared to larger block sizes, achieving competitive results against leading autoregressive models like Qwen3-8B on mathematical and code reasoning benchmarks. This advancement provides a practical framework for developing efficient diffusion models capable of complex reasoning tasks.