Unverified50% confidenceBenchmarkExact time
在相同测试任务中，DeepSeek V4 Pro预览版耗时约5分30秒但未能成功修复Bug
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
9/5/2026
First Seen
Valid until: 9/19/2026
Sources
Related Entities
Related Claims
UnverifiedDeepSeek V4 Pro在Humanity's Last Exam基准测试中得分37.769% similarUnverifiedDeepSeek-V3发布时在多项基准测试中与Claude 3.5 Sonnet持平甚至超越，训练成本据报道仅为主流闭源模型的数分之一61% similarUnverified在Claude Code环境中使用V4 Pro配合Thinking推理模式进行T5测试，耗时约6分48秒，生成的页面存在元素脚部出现在左上角的位置错乱Bug60% similarUnverified根据Commander Code官方内部评测数据，DeepSeek V4 Pro在高难度工具依赖任务中10次有6次击败了Claude Opus60% similarUnverified在Agent's Last Exam、ZeroBench、DeepSWE三项测试中，V4-Flash-Vision-Exp成绩反超Opus 4.859% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/858917API
curl https://kongchang.com/api/v1/knowledge/claims/858917MCP
get_claim(id=858917)