gpt-6IMAGE
GPT-6 Astra 幻觉率对比数据引讨论
一张基准图显示幻觉率：Argon 15.1%、GPT-6 Astra 44.8%、Claude Fable 5.1 72.6%，发帖者质疑为何没人关注这项指标。
bidyut
@bidyut_io29 天前收录原推发布于 2026.10.05
bidyut @bidyut_io原推发布于 2026.10.05
Wait… yes, we know 1M output tokens matters. But considering Gemini’s history with hallucinations, this might be the more important benchmark. Argon: 15.1% GPT-6 Astra: 44.8% Claude Fable 5.1: 72.6% Not context. Not tokens. Why isn’t everyone talking about this?
这条是引用着另一条写的：x.com/bidyut_io/status/2105501411291386143
x.com/bidyut_io/status/2107079726334714113采集于 2026.10.06 10:04