A table of 1,200 model benchmarks since 2019, updated daily (mostlyright.md)

Epoch AI has launched a comprehensive benchmarking hub, now featuring a vast collection of 1,200 large language model (LLM) benchmarks updated daily. This extensive database encompasses results from 710 models, including those developed by prominent organizations like OpenAI, Google DeepMind, and Meta, dating back to October 2019. Users can access over 6,800 individual results for various benchmarks, including GPQA Diamond and Humanity's Last Exam, enabling comparisons of model performance across multiple evaluations and tracking how benchmark scores evolve with each new model release. The significance of this hub lies in its ability to enhance transparency and accessibility in AI model evaluation, crucial for researchers and developers in the AI/ML community. By providing detailed metrics—including training compute estimates, result units, and evaluation setups—the platform aids in understanding the performance dynamics of different models, ensuring more informed decisions in model development and deployment. Moreover, the automatic update feature allows real-time tracking of new benchmarks and scores, establishing a robust resource for performance metrics in an ever-advancing field.