Unverified50% confidenceFactExact time
A 2023 study found that mainstream code models exhibit a hallucination rate of 15–25% when generating API calls, which is especially prominent when involving newer or more obscure libraries.
1
Sources
50%
Confidence
Long-term
Relevance
8/10/2026
First Seen
Sources
AI模型GitHub连接器可靠性实测：Kimi与Perplexity谁更值得信赖
Related Claims
UnverifiedResearch shows that structured prompts, compared to casual descriptions, can improve the first-pass rate of code generation from 30% to over 70%69% similarUnverified早期模型在某些领域的事实幻觉率高达15-25%，尤其在长尾知识领域幻觉率更高69% similarUnverified2024 年的多项基准测试表明，经过严格安全对齐的模型在代码生成任务上的拒绝率比基座模型高出 15-30%，在创意写作任务上的自我审查率高达 40%67% similarUnverified有研究表明当输入文本具有明确结构化标记时，主流LLM的信息提取准确率可提升20%以上，幻觉发生率相应降低66% similarVerifiedResearch shows that even the most advanced models have hallucination rates in factual question-answering tasks that still fluctuate between 5% and 20%.65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/722057API
curl https://kongchang.com/api/v1/knowledge/claims/722057MCP
get_claim(id=722057)