Event ·Multi-source

Jev 路由实测对比 GPT-6 Astra：公开基准数字参考价值有限

独立研究者用自建 harness 测试 Jev Router 在智能体任务中的表现，成绩与 GPT-6 Astra low 相当，但运行时间更长。Theo 用另一套 scoped tests 测了 DeepSWE，两套测试内容不同，结论有分歧也有共识。作者强调 Jev 是 System One 模型，不应与 GPT-6 Astra 这类 System Two 模型直接对比，更值得探索的是两者结合的路由方案。

Current read

有人拿自建 harness 实测了 Jev 路由，性能能追平 GPT-6 Astra low，代价是跑得更久。文章还讨论了为什么公开基准分数别太当真。

11 sources72° AI heatLast updated 2026/9/26 21:13:12

Evidence

11 sources

Primary sourceelvis

2026/9/26 21:13:12View source

Related source 1Artificial Analysis

2026/9/24 23:30:55View source

Related source 2@OpenAIDevs

2026/9/25 16:54:29View source

Related source 3Y Combinator

2026/9/26 14:35:20View source

Related source 4Decoder

2026/9/27 10:59:11View source

Related source 5arXiv cs.AI

2026/9/28 13:28:17View source

Related source 6kimmonismus

2026/9/28 17:44:10View source

Related source 7IT之家

2026/9/25 03:40:18View source

Related source 8Viking

2026/9/25 14:10:49View source

Related source 9Aravind Srinivas

2026/9/25 14:24:21View source

Related source 10Justine Moore

2026/9/25 18:14:05View 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