Year: 2026
Type: preprint
Abstract: Benchmark scores are a central currency in model releases: they inform purchasing decisions, shape public trust, and influence policy. Yet, a key assumption underlying benchmark scores is that the model performance measured through APIs faithfully reflects the behavior of deployed systems. We challenge this assumption by auditing ChatGPT, Claude, and Gemini across seven systems and nine benchmarks spanning general capability, social bias, and sycophancy. We find systematic API--interface differences in both accuracy and consistency. On average, API evaluations score 3.4 percentage points higher in accuracy and 2.1 percentage points higher in test--retest agreement than corresponding interface evaluations. For ChatGPT, the performance difference between API and interface access rivals the API-only difference between GPT 5.3 and GPT 5.4. Put differently, switching access surfaces can degrade performance as much as downgrading a full model version. We further test whether exposed API controls can reproduce interface behavior by varying system prompts, sampling parameters, and reasoning settings. These controls shift behavior in some cases but do not reliably eliminate the gap. Our findings document a context-validity gap: measurements obtained through APIs do not necessarily generalize to corresponding deployed interfaces, complicating the use of API evaluations as proxies for deployed systems.
Source: arXiv (Cornell University)
Institution: Stanford University
Subfield: Safety Research
Field: Social Sciences
Domain: Social Sciences
Open Access status: green
Awards: Collaborative Research: SCH: Fair Federated Representation Learning for Breast Cancer Risk Scoring, CAREER: Probabilistic Models for Spatiotemporal Data with Applications to Dynamic Brain Connectivity, Large Language Model Society: Opportunities and Risks for Information Seeking, Collaborative Research: III: Medium: Advancing Large Language Model Unlearning: Foundations and Applications
OpenAlex ID: W7212082014
Cites:
Cited by:
Related to: