Unverified50% confidenceOpinionExact time
传统代码基准如 HumanEval 和 MBPP 主要考察独立封闭问题，与真实开发场景存在较大差距
1
Sources
50%
Confidence
Long-term
Relevance
7/11/2026
First Seen
Sources
Claude Sonnet 4重回Cursor：CursorBench登顶，成本最高如何抉择
twittercursor_ai7/1/2026
Related Claims
UnverifiedCurrent LLMs tend to generate seemingly reasonable but flawed code (hallucinations in the code domain) on tasks that exceed their capability boundaries, which developers lacking engineering experience find difficult to identify during code review.64% similarUnverifiedThe performance gap among mainstream frontier models on standard tasks such as general question answering, text summarization, and code generation is narrowing, a phenomenon referred to as capability homogenization.62% similarUnverifiedMulti-instance parallelism faces engineering challenges including code conflicts, context consistency, and inter-task dependency management.61% similarUnverifiedBDD has faced controversy in recent years, with critics pointing out that scenario maintenance can easily become decoupled from code implementation and that Step Definition reuse leads to fragile tests.60% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/486759API
curl https://kongchang.com/api/v1/knowledge/claims/486759MCP
get_claim(id=486759)