Event ·Multi-source

Nvidia 发布 SoL-Pi：优化 harness 让编程智能体 token 消耗减少近一半

Nvidia 的 SoL-Pi 系统通过优化模型与环境之间的控制层（harness），将编程智能体的 token 用量最多降低 49%，性能基本不变。开发过程中，一个研究智能体测试了 152 种方法、跑了超过 3000 次实验才得出这套方案。不过在 SWE-bench 之外的基准上，收益有所缩小。

Current read

Nvidia 搞了个 SoL-Pi，不改模型只优化 harness，编程智能体 token 省了 49%，跑 3000 多次实验试出来的。

8 sources73° AI heatLast updated 2026/9/26 10:30:32

Evidence

8 sources

Primary sourceDecoder

2026/9/26 10:30:32View source

Related source 1shao__meng

2026/9/28 02:45:06View source

Related source 2n8n

2026/9/25 11:00:19View source

Related source 3arXiv cs.LG

2026/9/25 17:20:22View source

Related source 4rohanpaul_ai

2026/9/26 06:41:17View source

Related source 5IT之家

2026/9/26 09:00:37View source

Related source 6NVIDIA

2026/9/28 09:17:16View source

Related source 7Tech in Asia

2026/9/28 09:47:12View 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