The field of large language models (LLMs) is rapidly advancing, with a focus on improving test-time scaling to enhance reasoning capabilities. Recent developments have shifted towards selective resource allocation, mode-conditioning, and adaptive inference methods to optimize performance. These innovations aim to address fundamental limitations in current approaches, such as uniform resource distribution and diversity collapse. Noteworthy papers in this area include: SCALE, which proposes a framework for selective resource allocation based on sub-problem difficulty, achieving substantial performance improvements with superior resource utilization. Mode-Conditioning Unlocks Superior Test-Time Scaling, which introduces the mode-conditioning framework to explicitly allocate test-time compute across reasoning modes, consistently improving scaling across controlled graph-search tasks and large-scale reasoning benchmarks. ZIP-RC, an adaptive inference method that equips models with zero-overhead inference-time predictions of reward and cost, improving accuracy by up to 12% over majority voting at equal or lower average cost. OptPO, a principled framework that adaptively allocates inference budgets, dynamically halting sampling once the posterior confidence in a consensus answer exceeds a specified threshold, reducing rollout overhead while preserving or improving accuracy. Model Whisper, which introduces Test-Time Steering Vectors (TTSV) to efficiently unlock the powerful reasoning potential of LLMs for specific tasks or new distributions, achieving significant relative performance gains on various models.