Unverified50% confidenceBenchmarkExact time
SWE-bench要求模型在真实GitHub仓库中定位问题并生成可通过CI测试的补丁，以Resolve Rate作为核心指标
1
Sources
50%
Confidence
Long-term
Relevance
7/19/2026
First Seen
Sources
Related Claims
UnverifiedSWE-bench requires models to resolve real software defects on GitHub and is the authoritative benchmark for evaluating agent-level programming capabilities.81% similarVerifiedSWE-bench uses real GitHub Issues directly as test problems, requiring models to locate and fix bugs within a complete code repository.81% similarUnverifiedSWE-bench requires models to resolve real GitHub issues, and the pass rate does not exceed 50% even for the strongest models.80% similarUnverifiedSWE-bench requires models to autonomously complete code modifications starting from real GitHub issues and pass test cases, making it closer to real-world engineering scenarios than HumanEval.77% similarVerifiedSWE-bench requires models to locate and fix actual issues in real GitHub repositories.77% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/558663API
curl https://kongchang.com/api/v1/knowledge/claims/558663MCP
get_claim(id=558663)