Is there "Secret Sauce'' in Large Language Model Development?
This addresses the debate on AI leadership and capability diffusion by showing compute dominates frontier advances, but proprietary techniques matter for efficiency elsewhere, with incremental analysis of existing data.
The study analyzed 809 LLMs to determine if proprietary techniques or compute scaling drive performance, finding that at the frontier, 80-90% of performance differences are due to compute, while away from the frontier, proprietary methods reduce compute needs by up to 40x efficiency differences within companies.
Do leading LLM developers possess a proprietary ``secret sauce'', or is LLM performance driven by scaling up compute? Using training and benchmark data for 809 models released between 2022 and 2025, we estimate scaling-law regressions with release-date and developer fixed effects. We find clear evidence of developer-specific efficiency advantages, but their importance depends on where models lie in the performance distribution. At the frontier, 80-90% of performance differences are explained by higher training compute, implying that scale--not proprietary technology--drives frontier advances. Away from the frontier, however, proprietary techniques and shared algorithmic progress substantially reduce the compute required to reach fixed capability thresholds. Some companies can systematically produce smaller models more efficiently. Strikingly, we also find substantial variation of model efficiency within companies; a firm can train two models with more than 40x compute efficiency difference. We also discuss the implications for AI leadership and capability diffusion.