Google DeepMind has launched a new large language model (LLM) named FLAMe-RM-24B, which has outperformed other models such as GPT-4 and Anthropic on RewardBench, a benchmark for evaluating generative models. FLAMe-RM-24B, trained on extensive human evaluations, achieved an overall score of 87.8% in 2024, the highest on AI safety among its peers. The model is part of an ongoing effort to enhance automatic evaluation and performance of LLMs, addressing challenges like high costs and citation accuracy. Concurrently, other research efforts, such as the NeedleBench framework, are evaluating the long-context retrieval and reasoning capabilities of LLMs, introducing tasks like the Ancestral Trace Challenge to assess their performance across various text lengths.