Among the hardest tasks for humans are those found in competitive programming where problems require sophisticated algorithmic thinking, puzzle solving, and the creation of effective code. As a domain to assess language models (LMs), it has not received enough attention, though. This study presents the ICPC benchmark, which consists of 254 international collegiate programming contest (ICPC) tasks. Each problem includes official analysis, reference code, and sample and high-quality unit and hidden tests. We are able to develop and evaluate a variety of LM inference techniques for competitive programming with these resources. With zero-shot chain-of-thought prompting, we find that o1 only achieves a 19.1% pass@1 solve rate. With our best inference technique, which combines muti-turn self-judge with reflection and retrieval over episodic information, raises this to 42.2%. Furthermore, we conduct a new human-in-the-loop investigation to gain a deeper understanding of the remaining difficulties. Surprisingly, we discover that o1 can solve 17 out of 18 problems that were previously unsolvable by any model or technique with just a few specific instructions. A footstep toward LMs with grounded, imaginative, and algorithmic thinking is provided by our quantitative findings and qualitative research. We open source our code at https://github.com/kraritt/zolve.

The fast growth of large language models (LLMs) necessitates the urgent need for new NLP benchmarks. We provide B-REASO, the first inclusive Bengali assessment suite created to evaluate advanced foundation model knowledge and reasoning skills in a Bengali language setup. The B-REASO includes multiple-choice questions with four different degrees of difficulty: professional, college, high school, and middle school. The questions cover 50 different fields, from science and engineering to the humanities. Alongside B-REASO, there is B-REASO HEAVY, a subset of extremely difficult B-REASO topics that need for sophisticated reasoning skills to answer. We do a thorough assessment of the most sophisticated LLMs on B-REASO, encompassing models with an English focus. Findings show that only Claude-3.5-Sonnet was able to get an average accuracy of more than 65%, indicating that contemporary LLMs still have a long way to go. We hope that B-REASO will support the creation and expansion of foundation models for Bengali users by assisting in the analysis of significant advantages and disadvantages of these models. We open-source our code and data at https://github.com/kraritt/b-reaso.