Basaltlabs Monolith-1.0 – #1 on Last Exam, AIME, GPQA Diamond, MMLU-Pro (basaltlabs.org)

Basaltlabs has unveiled Monolith-1.0, a new AI model that has achieved the top performance on several prominent reasoning benchmarks, including Last Exam, AIME, GPQA Diamond, and MMLU-Pro. This distinction highlights Monolith-1.0's advanced capabilities in reasoning, mathematics, and science, coupled with its proficiency in long-context analysis. The model's extensive evaluation protocols and reproduction details are clearly outlined, ensuring transparency and reliability for researchers aiming to replicate the results. The significance of Monolith-1.0 for the AI/ML community lies in its comprehensive performance across multiple critical domains, which sets a new standard for future AI models. By making the weights available on Hugging Face alongside detailed evaluations and training information, Basaltlabs fosters collaboration and accelerates innovation within the field. This commitment to transparency not only enhances the reproducibility of AI research but also paves the way for subsequent advancements in reasoning and contextual understanding in AI applications.