Unverified50% confidenceBenchmarkExact time
DeepSeek V4 Pro的代码工程62.7分比Fable 5低7.3分，数据科学难级67.2分落后Opus 4.8约4.5分，高难推理60分位居全球第二
1
Sources
50%
Confidence
Short-term (~14 days)
Relevance
8/19/2026
First Seen
Valid until: 9/2/2026
Sources
Related Claims
VerifiedDeepSeek-V3在多项评测中性能超越了通义千问2.5和Llama 3.1等开源模型77% similarUnverifiedDeepSeek V4 Pro的软件工程能力从预览版的12.8分提升到62.7分，网络安全达到83.3分77% similarUnverifiedDeepSeek V4 Pro在本次测评中总分691分，排名第一75% similarUnverifiedDeepSeek的编程智能体DeepSWE在SWE-bench基准测试中从预览版的12.8分提升至62.7分75% similarUnverifiedDeepSeek-V4-Pro截至目前尚未超越顶级竞品的跑分水平，部分基准分数仍属预估值74% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/774250API
curl https://kongchang.com/api/v1/knowledge/claims/774250MCP
get_claim(id=774250)