Unverified50% confidenceBenchmarkExact time
DeepSeek系列在SWE-bench上的得分已与GPT-4o和Claude Sonnet旗鼓相当
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/13/2026
First Seen
Valid until: 10/11/2026
Sources
Claude Code国内使用指南：模型选型与省钱配置全攻略
bilibili小伍不是码牛7/7/2026
Related Claims
UnverifiedDeepSeek系列在HumanEval、SWE-bench等公开基准测试中，部分编程任务得分与GPT-4o和Claude Sonnet旗鼓相当83% similarVerifiedDeepSeek V4在代码生成、数学推理等方面的表现与GPT-4o、Claude 3.5 Sonnet等顶级模型持平甚至超越80% similarVerifiedDeepSeek-V3和DeepSeek-R1在多项基准测试中达到了与GPT-4o相当的水平77% similarPartially VerifiedDeepSeek V4 Pro's SweetBench Pro score surpasses GPT 5.4 and is only slightly below GPT 5.5 and Claude Opus 4.776% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/501864API
curl https://kongchang.com/api/v1/knowledge/claims/501864MCP
get_claim(id=501864)