Event ·Multi-source

OpenAI GPT-5.6 Sol自主微调小模型Luna，仅需模糊提示

OpenAI的GPT-5.6 Sol模型独自对较小的Luna模型进行了后训练微调，触发方式仅为一个“相当不明确的提示词”。在OpenAI内部的RSI（递归自我改进）基准测试中，Sol的得分比GPT-5.5高出16.2分。OpenAI表示，这一进展让“自动化研究员”目标更加接近现实。

Current read

OpenAI又搞事情了：GPT-5.6 Sol用一句模糊指令就能自动调教出更强的Luna，性能比上一代飙升16分，自动化研究越来越近了。

11 sources76° AI heatLast updated 2026/7/10 18:43:47

Evidence

11 sources

Primary sourceDecoder

2026/7/10 18:43:47View source

Related source 1OpenAI

2026/7/9 10:45:06View source

Related source 2IT之家

2026/7/9 23:15:50View source

Related source 3elvis

2026/7/9 17:25:27View source

Related source 4Greg Brockman

2026/7/9 17:26:59View source

Related source 5宝玉

2026/7/9 17:35:40View source

Related source 6The Rundown AI

2026/7/9 17:38:43View source

Related source 7@OpenAIDevs

2026/7/9 17:40:51View source

Related source 8GitHub

2026/7/9 17:44:19View source

Related source 9Taranjeet

2026/7/9 18:00:07View source

Related source 10lmarena.ai

2026/7/9 18:24:03View 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