Unverified50% confidenceEventExact time
Across 120 evaluations, the only configuration that answered all questions correctly (24/24) was the OpenAI cached mode (Responses API's web_search plus external_web_access: false).
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/16/2026
First Seen
Valid until: 1/14/2027
Sources
Related Entities
Related Claims
VerifiedWhen concurrent requests exceed a threshold, the OpenAI API returns error code 429.65% similarUnverifiedMost platforms that follow the OpenAI API-compatible format include a usage object in their responses, usually carried in the last event of a streaming response with complete token statistics65% similarUnverifiedOpenAI在每次API响应的HTTP头中返回x-request-id字段，作为全局唯一标识符用于定位特定请求的处理记录64% similarUnverifiedOpenAI-compatible interfaces usually cover only the core fields, and implementations of extended capabilities such as tools, response_format, and streaming events are not fully consistent across vendors; when encountering unsupported features, they may return a parameter error or silently ignore it.63% similarUnverifiedAnthropic要求至少1024个token的前缀才会触发缓存，OpenAI的缓存阈值为128个token62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/1045021API
curl https://kongchang.com/api/v1/knowledge/claims/1045021MCP
get_claim(id=1045021)