BREAKING
Cognition Ships SWE-1.7
FrontierCode 1.1 Main scores
Opus 4.846.5
GPT-5.543
SWE-1.742.3
Kimi K2.730.1
0%
Terminal-Bench 2.1
0%
SWE-Bench Multiling.
0$
Avg cost per task
0TPS
Cerebras inference
0%
Weight delta compression
Performance vs trustworthiness
Cost edgePareto
Frontier performance
Fraction of the cost
Undercuts rivals
Trust focusSafety
Kimi K2.7 concerns
Refuses privacy misuse
Matches US models
Free for paid users, briefly
AI NEWS BLITZ
Cognition has released SWE-1.7, a coding model built on China's Kimi K2.7.