As artificial intelligence’s capabilities improve, it is increasingly viewed as a general scientific method. But how true are these claims? Does AI outperform all techniques, or only some, and how is this changing? To assess the claims, we assemble a corpus of 2,507 head-to-head comparisons between AI and other scientific analysis techniques across 27 scientific disciplines from papers published between 2000 and early 2025. We find a profound dichotomy. Relative to traditional statistics, AI often outperforms, but at a significantly higher computational cost. But there are also nearly a quarter of cases where AI is both more expensive and performs worse than traditional statistical techniques and this fraction has been stable for a decade. Relative to scientific computing, AI often underperforms, but at lower computational cost. This has begun to change: since 2020, AI’s performance against scientific computing has notably strengthened and it now outperforms on more than half of comparisons. These patterns suggest that AI is therefore not a universal replacement for existing methods, but rather a valuable – and improving – part of a new AI-enabled scientific frontier.

We characterize the lifecycles of open-weight AI models using a weekly panel of the 108,074 mostdownloaded models on Hugging Face, which together account for 99.5 percent of the platform's more than 53 billion downloads. We link these data to token usage from OpenRouter and performance rankings from Arena and Artificial Analysis. We establish key institutional details about the supply of and demand for AI models, and we organize our analysis into three phases: birth, life, and death. In the "birth" phase, we show that major developers often release models in batches, concurrently introducing models that are both vertically differentiated (e.g., by size, capability, and latency) and horizontally differentiated (e.g., by task specialization, hardware compatibility, and alignment). Because weights are open, developers can also build on one another's models. We document cumulative innovation through fine-tuned, quantized, merged and adapted descendants of source models. In the "life" phase, we quantify how demand evolves after release. Even within the sample of top models, demand is concentrated, and the concentration persists: the median model loses ∼68 percent of its release-week downloads within ten weeks, while the most popular models decline gradually; as a result, the top percentile's advantage over the median model is ∼600-fold at release, and ∼900-fold six months later. Further, we estimate spillover effects from new popular model releases on related models and find that the first popular third-party descendant is associated with a 76 to 99 percent increase in weekly downloads of its source model. In the "death" phase, we construct measures of technical obsolescence and usage obsolescence. We find that technical obsolescence is associated with a slow but persistent decrease in demand, while usage obsolescence coincides with an immediate but temporary drop in demand. Our findings characterize open-weight AI as a market of complementary model portfolios, one in which third-party developers building on a source model can raise demand for the original, and in which demand is slow to respond to technical obsolescence.

Scaling laws describe how language model capabilities grow with compute and data, but say nothing about how long a model matters once released. We introduce time-to-peak and lifespan as measures of model obsolescence and use them to characterize the scientific adoption trajectories of 62 LLMs across more than 108k citing papers (2019-2025), separating active adoption from background citation to recover per-model trajectories that citation counts cannot resolve. We find that a model's longevity is shaped more by when it was released than by its characteristics: release year predicts time-to-peak and lifespan more strongly than architecture, openness, or scale. LLM adoption follows an inverted-U curve (rising after release, peaking, and then declining), but this pattern is rapidly compressing. Each successive release year is associated with a 27% shorter time-to-peak and a 23% shorter lifespan (p<0.001), robust to minimum-age thresholds and controls for model size. These adoption-side dynamics are invisible to scaling laws and suggest that specialization on any single model may be a depreciating investment, with costs falling on reproducibility and migration.