Expired50% confidenceOpinionExact time
In code generation benchmarks such as SWE-bench and HumanEval, domestic models still lag behind Claude 3.5/4, GPT-4, and others, with the gap being especially pronounced on high-difficulty tasks such as complex multi-file refactoring and cross-language code conversion.
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
7/2/2026
First Seen
Valid until: 9/30/2026(expired)
Sources
Kimi Code Open-Source Hands-On Review: Installation, Feature Comparison & Cost Analysis
bilibili鲲鹏Talk6/9/2026
Related Claims
UnverifiedGPT-5、Grok 4等模型在基准测试（如HumanEval、SWE-bench）上表现优异，但这些测试通常是单次代码生成，与需要多轮工具调用的真实工程任务存在显著差距80% similarUnverifiedGPT-4、Claude等模型的能力边界覆盖了文字生成、代码编写、信息综合等典型白领入门级任务77% similarUnverifiedGPT-4、Claude等大语言模型在预训练阶段消化了海量软件工程文档、测试规范和历史缺陷报告，天然理解等价类划分、边界值分析等经典测试设计方法77% similarUnverifiedThe Claude series has demonstrated capabilities on par with or even surpassing GPT-4 across multiple benchmarks, with particularly notable advantages in scenarios such as code generation and long-text understanding.77% similarUnverifiedOpen-source models such as Meta's Llama 3.1 series, DeepSeek-V3, and Alibaba's Qwen2.5-Coder have significantly narrowed the gap with GPT-4-level models on code generation benchmarks such as HumanEval and SWE-bench.76% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/46886API
curl https://kongchang.com/api/v1/knowledge/claims/46886MCP
get_claim(id=46886)