Lightning Weave: Improving the Accuracy-Efficiency Frontier of Reasoning Models through Capability Composition

Benchmark Model Rank Results
mathematical-reasoning-on-aime-2025Lightning Weave (Qwen3.5-4B)–Accuracy: 74.6
mathematical-reasoning-on-aime24Lightning Weave (Qwen3.5-4B)–Acc: 84.2
mathematical-reasoning-on-hmmt-feb-2025Lightning Weave (Qwen3.5-4B)–Accuracy: 64