Unverified75% confidenceOpinionTime unknown
测试者建议涉及代码生成、逻辑推理、空间理解的场景应使用Claude Sonnet 3.5或更高级别的模型，而非Claude Haiku 4.5
1
Sources
75%
Confidence
Medium-term (~90 days)
Relevance
6/3/2026
First Seen
Valid until: 9/1/2026
Sources
Claude Haiku 4.5实测：5项编程任务全面翻车
bilibilikarminski-牙医
Related Entities
Related Claims
Unverified在Pillow库RGB转HSV精度Bug测试中，Claude Code使用Sonnet 4.6模型一次通过，逻辑清晰，获得满分73% similarUnverified代码理解与生成能力强的模型（如Claude 3.5 Sonnet、GPT-4o）在处理混淆JS时通常表现更稳定68% similarUnverified一位B站内容创作者对Claude Haiku 4.5进行了系统性的视觉编程任务测试68% similarVerifiedClaude Code依托的Sonnet模型在HumanEval、SWE-bench等主流编程基准测试中持续领先67% similarUnverifiedClaude Sonnet 4模型在编程基准测试SWE-bench上表现优异，SWE-bench由普林斯顿大学开发67% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/40311API
curl https://kongchang.com/api/v1/knowledge/claims/40311MCP
get_claim(id=40311)