There’s a part of the 6.2 launch OpenAI doesn’t mention. In Codex, Astra 6.2 is 4.7x slower than Astra 6. Sol 6.2 is slower too: 72 tok/s on Fast vs 200 on 6.1. And the $200 plan still gives half the API value of Claude Max, while 6.2 loses to the old models on some benchmarks. So why make it slower? OpenAI says 6.2 needs far fewer tokens per task than Fable, which makes it cheaper to run for Op...

🌐 链接: https://x.com/miu21590/status/2122811388670157302
🏷️ 分组: X|正在关注
🕒 时间: 2026-11-18 05:57:29 vechen (@miu21590) on X