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 residual 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.

Germline genetic testing plays a critical role in diagnosing inherited predispositions and increasingly guides therapeutic and surveillance choices—but becomes technically challenging after allogeneic hematopoietic stem cell transplantation (HSCT), when donor-derived DNA contaminates host tissues. To address this, we compared donor-derived DNA across three accessible tissues—buccal swab, nail, and eyebrow follicles—in recipients after hematopoietic stem cell transplantation using two orthogonal assays (34-SNP next-generation sequencing and a 27-marker short tandem repeat panel) and modeled clinical covariates that influence chimerism. Eyebrow follicles showed consistently low donor DNA (median 1% by NGS; 3% by STR) whereas buccal swabs and nails carried substantially higher donor fractions (+25 and +22 percentage points versus eyebrow, respectively; both p < 0.01). Across methods, STR yielded on average ≈6 percentage points higher donor fractions than NGS at low-level chimerism. Several transplant covariates correlated with chimerism: matched-related donors and a perfect HLA match (10/10) were each associated with lower donor DNA (≈12–14 and 15–20 percentage points, respectively); longer times since hematopoietic stem cell transplantation correlated with lower levels for nail samples, and donor–recipient sex match correlated with higher donor DNA (~7–8 percentage points). Even low-level chimerism can distort germline variant interpretation. We propose a pragmatic protocol for post-hematopoietic stem cell transplantation germline testing that prioritizes eyebrow follicles as the default tissue. An SNP-based quality control assay is used to flag unsafe donor fractions (≥ 5–10%) before comprehensive germline analysis, reducing the risk that chimeric donor DNA distorts germline variant interpretation.