Test-Time Scaling with Diffusion Language Models via Reward-Guided Stitching

Researchers developed a new framework called 'Stitching Noisy Diffusion Thoughts' that improves AI reasoning by combining the best parts of multiple solution attempts rather than just selecting complete answers. The method achieves up to 23.8% accuracy improvement on math and coding tasks while reducing computation time by 1.8x compared to existing approaches.

• →New framework combines step-level components from multiple AI reasoning attempts rather than selecting entire solution paths.
• →Uses diffusion sampling for exploration, reward models for scoring, and autoregressive models for final answer synthesis.
• →Achieves up to 23.8% accuracy improvement across six math and coding benchmarks.
• →Provides 1.8x latency reduction compared to traditional diffusion models and unified architectures.
• →Framework is training-free and shows greatest benefits on harder reasoning problems.