Event ·Multi-source

OpenAI 多次修改 GPT-6 Astra 基准测试数据

OpenAI 在发布 GPT-6 Astra 模型公告后多次修改评测基准数据。Astra 幻觉率从 4.2% 降至 2%后又恢复至 4.2%。Anthropic Fable 5.1 在 FrontierMath Tier 4 评测中成绩从 87.8% 降至 78%后回升至 83%。OpenAI 表示数据修正代表模型可用性能的最佳估计。

Current read

OpenAI 发布 GPT-6 Astra 后多次调整评测数据，部分指标波动近10个百分点，引发对基准测试准确性的质疑。

11 sources73° AI heatLast updated 2026/9/6 06:46:16

Evidence

11 sources

Primary sourceIT之家

2026/9/6 06:46:16View source

Related source 1The Rundown AI

2026/9/4 18:58:06View source

Related source 2shao__meng

2026/9/7 06:38:22View source

Related source 3向阳乔木

2026/9/4 08:57:07View source

Related source 4Decoder

2026/9/4 11:07:36View source

Related source 5lmarena.ai

2026/9/4 20:57:58View source

Related source 6e27

2026/9/4 09:49:04View source

Related source 7Milvus

2026/9/4 15:52:00View source

Related source 8OpenAI

2026/9/4 20:13:05View source

Related source 9@OpenAIDevs

2026/9/4 20:13:27View source

Related source 10Sam Altman

2026/9/4 20:17:43View source

Conflict check

Dedup data only shows that these sources discuss the same event, not that they agree. There is no structured supporting or refuting evidence yet, so no conflict is inferred automatically.

Open the story