Unverified60% confidenceBenchmarkExact time
在SWE-Bench Multilingual测试中，HY3得分75.8，略微落后于DeepSeek V4 Pro的76.2
2
Sources
60%
Confidence
Medium-term (~90 days)
Relevance
7/13/2026
First Seen
Valid until: 10/11/2026
Sources
腾讯混元HY3深度评测：小体量开源模型如何逆袭大厂
bilibili硅谷神奇7/7/2026
Related Claims
Unverified在SWE-Bench Pro上，HY3以57.9分反超DeepSeek V4 Pro的55.4分74% similarUnverified在SWE-Bench评分中，o1的得分远高于o3-mini medium，但略低于o3-mini-high69% similarUnverifiedHy3在MRCR（多轮指代消解）评测中得分从42.9%跃升至75.1%66% similarUnverified通义千问3.6在SWE-Bench测试中得分78.8分，对比Claude Opus的80.9分66% similarUnverifiedQwen3-4B在多项推理类评测指标上已与DeepSeek V3或GPT-4o相差无几65% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/495158API
curl https://kongchang.com/api/v1/knowledge/claims/495158MCP
get_claim(id=495158)