Per-Response Model Identity Disclosure in Frontier AI APIs: Cross-Vendor Comparison Data (June 2026)

Controlled comparison of per-response model identity disclosure across Google Gemini API and Anthropic Claude API. Google returns model_version and x-gemini-service-tier in every response (8/8 tests). Anthropic returns no equivalent fields (0/4 tests); response.model echoes the requested model without confirming the served model. Demonstrates that routing transparency is technically feasible and its absence is a design choice.针对Google Gemini API与Anthropic Claude API的逐响应模型身份披露情况开展对照控制实验。在全部8项测试中，Google Gemini API会在每一次响应中返回model_version与x-gemini-service-tier字段（8/8 tests）；Anthropic Claude API在全部4项测试中均未返回等效字段（0/4 tests），其响应内的response.model字段仅回显用户请求的模型，未确认实际部署的模型。本实验表明，路由透明度（routing transparency）在技术层面具备实现可行性，而未提供该类信息实为一种设计选择。