Unverified80% confidenceFactExact time
o3-mini-high在SWE-Bench上修复了39%的错误，允许调用内部工具后成功率飙升到61%
1
Sources
80%
Confidence
Long-term
Relevance
2/5/2025
First Seen
Sources
Related Entities
Related Claims
UnverifiedCodex One's performance on SWE-bench is slightly better than O3 High, achieving an accuracy rate of 75% on OpenAI's internal software engineering tasks, compared to 70% for O3 High.64% similarUnverifiedIn the SWE-Bench score, o1 scored significantly higher than o3-mini medium but slightly lower than o3-mini-high.62% similarUnverifiedOn SWE-Bench Pro, HY3 surpassed DeepSeek V4 Pro with a score of 57.9 versus 55.4.61% similarUnverifiedStep 3.7 Flash 在 SWE-Bench PRO 基准测试中以 56.3 分排名第二60% similarUnverifiedThe resolution rate of the best-performing systems on the SWE-bench benchmark has risen from less than 5% initially to over 40%.59% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/658API
curl https://kongchang.com/api/v1/knowledge/claims/658MCP
get_claim(id=658)