Unverified50% confidenceBenchmarkExact time
Cognition表示SWE-2在其自有编程测试中的得分与Anthropic的Claude模型仅相差一分，成本却低64%
1
Sources
50%
Confidence
Medium-term (~90 days)
Relevance
10/8/2026
First Seen
Valid until: 1/6/2027
Sources
Related Entities
Related Claims
UnverifiedCognition发布编码模型Sway 2，在Frontier Code基准上取得50.0%，据称与更强模型差距在1分以内且成本低64%76% similarUnverified对于两个被测试的模型，开启最大推理能力后深度谜题上的让步率从19.2%和12.5%直接降到了0%70% similarUnverified研究版Claude自主攻关约一天半时间，将黎曼Zeta函数零点下界从41.6%提升到67.2%，并给出可形式化验证的结果69% similarUnverified完成相同任务时，SWE-2所需的交互轮次比前代SWE-1.7减少了58%，成本降低了81%，同时得分更高68% similarUnverified在Artificial Analysis智能指数综合测试中，GPT-5.6性能比Claude Opus 4.5略低约0.9分，但成本几乎减半68% similar
Cite This Claim
Stable URI
https://kongchang.com/claim/989082API
curl https://kongchang.com/api/v1/knowledge/claims/989082MCP
get_claim(id=989082)