{"paper":{"title":"API Benchmark Scores Do Not Reliably Transfer to Chatbot Interfaces","arxiv_id":"2609.08861","primary_cat":"cs.AI","submitted_at":"2026-09-08T15:08:37+00:00","version_compiled":"v2"},"notice":"Every quoted sentence is the authors' text from the arXiv listing abstract. Labels and probabilities are the outputs of classifier-1.13.0 on the question bank abstract-p1-20260923, exactly as returned. No sentence in this record was written by a language model.","schema":"pith.compiled-record/v0","status":"compiled","abstract":{"source":"arXiv listing abstract v2","sentences":[{"i":0,"p":1.0,"id":"s0","own":false,"role":"background","text":"Benchmark scores are a central currency in model releases: they inform purchasing decisions, shape public trust, and influence policy.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.04,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.0,"background":1.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":1,"p":0.81,"id":"s1","own":false,"role":"background","text":"Yet, a key assumption underlying benchmark scores is that the model performance measured through APIs faithfully reflects the behavior of deployed systems.","hedge":{"p":0.99,"label":"asserted"},"own_p":0.05,"main_claim_p":0.03,"quantitative":false,"quantitative_p":0.03,"role_probabilities":{"gap":0.19,"method":0.0,"result":0.0,"background":0.81,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":2,"p":0.75,"id":"s2","own":true,"role":"contribution","text":"We challenge this assumption by auditing ChatGPT, Claude, and Gemini across seven systems and nine benchmarks spanning general capability, social bias, and sycophancy.","hedge":{"p":0.96,"label":"asserted"},"own_p":0.95,"main_claim_p":0.04,"quantitative":true,"quantitative_p":0.76,"role_probabilities":{"gap":0.0,"method":0.25,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.75,"interpretation":0.0}},{"i":3,"p":1.0,"id":"s3","own":true,"role":"result","text":"We find systematic API--interface differences in both accuracy and consistency.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.96,"main_claim_p":0.25,"quantitative":false,"quantitative_p":0.13,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":4,"p":1.0,"id":"s4","own":true,"role":"result","text":"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.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.96,"main_claim_p":0.14,"quantitative":true,"quantitative_p":0.99,"role_probabilities":{"gap":0.0,"method":0.0,"result":1.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":5,"p":0.81,"id":"s5","own":true,"role":"result","text":"For ChatGPT, the performance difference between API and interface access rivals the API-only difference between GPT 5.3 and GPT 5.4.","hedge":{"p":1.0,"label":"asserted"},"own_p":0.93,"main_claim_p":0.07,"quantitative":true,"quantitative_p":0.97,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.81,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.19}},{"i":6,"p":0.94,"id":"s6","own":true,"role":"interpretation","text":"Put differently, switching access surfaces can degrade performance as much as downgrading a full model generation.","hedge":{"p":0.87,"label":"asserted"},"own_p":0.81,"main_claim_p":0.13,"quantitative":true,"quantitative_p":0.92,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.06,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.94}},{"i":7,"p":0.95,"id":"s7","own":true,"role":"method","text":"We further test whether exposed API controls can reproduce interface behavior by varying system prompts, sampling parameters, and reasoning settings.","hedge":{"p":0.74,"label":"asserted"},"own_p":0.95,"main_claim_p":0.02,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.0,"method":0.95,"result":0.0,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.05,"interpretation":0.0}},{"i":8,"p":0.99,"id":"s8","own":true,"role":"result","text":"These controls shift behavior in some cases but do not reliably eliminate the gap.","hedge":{"p":0.96,"label":"asserted"},"own_p":0.96,"main_claim_p":0.06,"quantitative":false,"quantitative_p":0.05,"role_probabilities":{"gap":0.0,"method":0.01,"result":0.99,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.0}},{"i":9,"p":0.99,"id":"s9","own":true,"role":"interpretation","text":"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.","hedge":{"p":0.61,"label":"asserted"},"own_p":0.94,"main_claim_p":0.96,"quantitative":false,"quantitative_p":0.06,"role_probabilities":{"gap":0.0,"method":0.0,"result":0.01,"background":0.0,"limitation":0.0,"future_work":0.0,"contribution":0.0,"interpretation":0.99}}]},"compiled":{"main_claim":{"p":1.0,"text":"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.","location":"abstract s9","hedge_strength":{"p":0.61,"label":"asserted"}},"provenance":{"route":{"via":"TypeSafe API","model":"classifier-1.13.0","endpoint":"https://api.typesafe.ai/v1/systemone","provider":"TypeSafe"},"packing":"one request per paper","requests":1,"input_tokens":7428,"cost_usd_reported":0.0},"compiled_at":"2026-09-23","bank_version":"abstract-p1-20260923","model_version":"classifier-1.13.0","abstract_states":{"claims_first":0.04,"states_limitation":0.36,"human_participants":0.09,"controlled_comparison":0.51,"randomized_assignment":0.02,"states_quantitative_result":0.99,"compares_with_prior_results":0.79,"mentions_code_or_data_availability":0.03},"evidential_basis":{"p":1.0,"value":"observation or experiment","probabilities":{"argument":0.0,"literature":0.0,"proof_or_derivation":0.0,"analysis_of_existing_data":0.0,"observation_or_experiment":1.0,"simulation_or_computation":0.0}},"contribution_type":{"p":1.0,"value":"empirical study","probabilities":{"review":0.0,"method_or_tool":0.0,"empirical_study":1.0,"simulation_study":0.0,"theoretical_result":0.0,"dataset_or_benchmark":0.0,"position_or_commentary":0.0}},"contribution_statement":{"p":1.0,"text":"We challenge this assumption by auditing ChatGPT, Claude, and Gemini across seven systems and nine benchmarks spanning general capability, social bias, and sycophancy.","location":"abstract s2"}},"coverage":{"tier":"abstract","levels_run":["P abstract"],"not_examined":["full text","figures and tables","references and citing sentences","relations to other papers"],"sources_read":["arXiv listing abstract v2"],"levels_not_run":["S sections","T body sentences","C citations","R cross-paper relations"],"sentences_examined":10,"sentences_in_abstract":10},"measured_accuracy":{"url":"https://pith.science/accuracy","measured_on":"2026-10-02","method":"two independent model families labelled a fresh random sample blind; the figure is how often the published label matches their consensus","sentence_role_agreement":0.92,"main_claim_agreement":0.908,"typed_citation_agreement":0.933,"typed_citation_agreement_95_interval":[0.917,0.945],"review_comments_accurate":0.758,"review_comments_accurate_95_interval":[0.684,0.82],"review_caught_admitted_error_found_or_related":0.448}}