Taskade 对 Opus 5.5、GPT-6 Sol / Luna 的早期实测

John Xie 谢青云 @johnxie原推发布于 2026.09.23

在 X 上打开原推

Claude Opus 5.5 just landed. GPT-6 Luna and Sol too. Early @Taskade TSK-1 benchmarks: Opus 5.5 scored 8/10. Sol 7/10. Luna 5/10. 243 builds in. Every build teaches TSK-1. We tune the harness and run again. Each model gets the same real jobs. One prompt. Build the app, data, agents, automations and logic. Then run it live and score the results. CRMs. Client portals. Intake systems. Fleet operations. AI governance. And more. All connected. Context. Memory. Tools. Data. Automations. How it all connects, and what happens after the first build. One prompt. One app. One living workspace.