Top 3 papers of the Week [Feb 9 - Feb 13, 2026] suggested by Genloop's LLM Research Hub.

🥇 Towards Autonomous Mathematics Research

The research introduces Aletheia, a generative AI agent for advanced mathematical reasoning, leveraging novel inference-time scaling laws, tool use, and human-AI collaboration to solve Olympiad-level to PhD-level problems, including autonomous solutions to open questions in mathematics.

Research Credits: Tony Feng; Trieu H. Trinh; G. Bingham; Dawsen Hwang; Yuri Chervonyi; Junehyuk Jung; Joonkyung Lee

Link

🥈 Step 3.5 Flash: Open Frontier-Level Intelligence with 11B Active Parameters

Step 3.5 Flash introduces a sparse Mixture-of-Experts model combining high-fidelity reasoning with efficient inference, leveraging novel attention mechanisms and scalable RL frameworks. It achieves frontier-level performance across coding, math, and agent tasks, redefining efficiency for deploying advanced agents in industrial environments.

Research Credits: Ang Li; Aobo Yang; Bin Wang; Binxing Jiao; Bojun Wang; Boyu Chen; Brian Li; Buyun Ma; Chang Su; Changxing Miao; Changyi Wan; Chao Lou; Chen Hu;

Link

🥉 OPUS: Towards Efficient and Principled Data Selection in Large Language Model Pre-training in Every Iteration

The research introduces OPUS, a dynamic data selection framework leveraging optimizer-induced update space for efficient pre-training of LLMs. It achieves superior performance and data efficiency across diverse corpora and model scales, outperforming industrial baselines and enabling advancements in specialized domains.

Research Credits: Shaobo Wang; Xuan Ouyang; Tianyi Xu; Yuzheng Hu; Jialin Liu; Guo Chen

Link