Unverified85% confidenceEventTime unknown
B站UP主通过5项可视化编程任务对Claude Haiku 4.5进行了系统性测试，结果全部表现不佳
1
Sources
85%
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测试者认为Claude Haiku 4.5在3D建模理解、代码性能优化、库引用准确性、物理逻辑推理、指令遵循能力五个维度均表现为差或极差81% similarUnverifiedClaude Haiku 4.5生成的大象牙膏动画代码存在严重性能瓶颈，运行时特别卡顿68% similarUnverified本次评测由B站UP主'不正经的前端'设计，包含四个难度递进的编码场景67% similarUnverifiedHaiku 4.5在安全评估中的错位行为（misalignment behavior）发生率在所有模型中最低。65% similarUnverified测试中约1-2B的小模型无法在大段系统提示词中准确定位并完成复杂工具调用任务62% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/40309API
curl https://kongchang.com/api/v1/knowledge/claims/40309MCP
get_claim(id=40309)