Unverified50% confidenceBenchmarkExact time
在使用同一checkpoint、同为max effort设置时，DeepSeek精简版harness在DeepStop UE上得分72.6，而OpenCode仅为65.5
1
Sources
50%
Confidence
Long-term
Relevance
9/18/2026
First Seen
Sources
Related Entities
Related Claims
UnverifiedDeepSeek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76% similarUnverifiedDeepStop UE（DeepSeek Open Source Benchmark for Unified Evaluation）是DeepSeek用于评估编程智能体能力的基准测试集，考察智能体与harness协同运作的整体表现72% similarUnverifiedDeepSeek V4.1 Flash在Deep SWE软件工程代码测试上拿到74分，是UP主见过的第一个在Deep SWE上达到74分的开放权重模型71% similarUnverifiedDeepSeek-R1在SWE-Bench Verified基准测试中的解决率为49.2%70% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/932802API
curl https://kongchang.com/api/v1/knowledge/claims/932802MCP
get_claim(id=932802)