A recent analysis examined the tool lists submitted to three AI APIs: Claude Haiku 5.5, GPT-5 nano, and Gemini 2.5-Flash. The study found that each API enforces its own unique rules for accepting tool lists, leading to varying refusal rates. Anthropic's API refused 456 lists, OpenAI's refused 591, and Google's refused 7, with an additional 135 unexplained refusals. The findings suggest that server authors should adhere to specific naming conventions, uniqueness, schema types, and tool count limits to ensure broader API compatibility. AI
IMPACT Developers need to be aware of differing API rules for tool lists to ensure broader compatibility and avoid refusals.
RANK_REASON The item details a study and its findings regarding the behavior of AI APIs. [lever_c_demoted from research: ic=1 ai=1.0]
AI-generated summary · Google Gemini · from 1 sources. How we write summaries →