Recent evaluations of AI models have highlighted significant advancements in multimodal reasoning capabilities. OpenAI's new model, referred to as 'o1', has demonstrated superior performance compared to Anthropic's Claude 3.5 Sonnet. Specifically, o1 has shown to be twice as effective as Claude 3.5 Sonnet with direct prompting and 10% better when utilizing self-debugging techniques for science agents. However, o1's operational costs are notably higher, being ten times more expensive than its competitors. The Claude 3.5 Sonnet has also received praise for its improved mathematical reasoning abilities, with users noting its capacity to engage in complex tasks and acknowledge limitations. Despite these advancements, some evaluations suggest that o1 still does not fully leverage domain knowledge, which may hinder its overall performance. The latest updates and benchmarks, including the newly released ScienceAgentBench, provide a clearer picture of the competitive landscape among AI models in the field of multimodal reasoning.