The team conducted 153 autonomous runs across 18 frontier AI models using the NanoGPT optimizer speedrun benchmark. The benchmark evaluates model performance on validation sets while tracking compute resources. Results varied by model, with reported validation accuracies ranging from 26.8% to 52.2% and compute usage measured in megabytes to billions of parameters. The highest reported accuracy was 52.2% for the model identified as "Opus" with 3,058 compute units, while the lowest was 26.8% for "Fable" with 2,726 compute units. Additional models showed intermediate values, with most using between 1.2 and 4.6 billion parameters. The study provides a standardized comparison of frontier models and may inform future optimizer design, though further validation is needed.