[NS : roy] SWE-bench榜:Claude Opus 5实测96%,前三全是Anthropic

看了眼SWE-benchVerified榜,ClaudeOpus5实测96%前三全是Anthropic,第四是Fireworks的Ember-1开源GLM-5停在77.8%,差了快20个点
 
 
Back to Top