AI编码
x.com ·
@omarsar0
NVIDIA 自演化 agent harness 论文：SoL-Pi 将 token 流量削减近半
X 用户转述 NVIDIA 关于自演化 agent harness 的论文，提出 SoL-Pi 方法：在 harness 层跨多个仓库与验证器环境运行自动研究循环，仅保留通过筛选的机制，最终保留 Action Fusion、Online Context Compact、ObservationPack、Evidence-Preserving Reducer 四项机制。帖子称其在 GPT-5.6 Sol 与 Opus 5 上匹配基线 harness，token 流量减少近半，在 51 任务 EdgeBench 上 API 成本约降三分之一，估算每小时节省 8.75–13.50 美元（对比原生 Codex 与 Claude Code harness）及 4.36–5.71 美元（对比基线 harness）。论文编号 arXiv 2609.20519，代码在 NVlabs 的 GitHub。上述数据为帖子自述，尚未经独立核实。
AI推荐理由 该帖转述 NVIDIA 一篇关于自演化 agent harness 的论文（arXiv 2609.20519），给出具体机制名称（Action Fusion、Online Context Compact、ObservationPack、Evidence-Preserving Reducer）和量化结果：token 流量约减半、51 任务 EdgeBench 上 API 成本约降三分之一，并给出每小时成本区间。信息密度较高且指向可查论文与 NVlabs 代码，但内容为第三方转述，未提供独立验证，成本与基准对比数据仅来自帖子自述，需以论文原文核实；互动量仅作传播信号，未计入质量评分。
72
浏览 80,873
点赞 1,674
评论 65
转发 167
正文
Build your own harness, folks.
This is absolute banger paper from NVIDIA on self-evolving agent harnesses.
(bookmark it)
They introduce SoL-Pi which cuts token traffic by nearly half.
And it matches its baseline harness on GPT-5.6 Sol and Opus 5.
More details below:
Instead of tuning a harness by hand, they run auto-research loops at the harness layer across many repository-derived and verifier-driven environments, keeping only the mechanisms that survive selection.
Four mechanisms survived:
> Action Fusion changes how actions execute
> Online Context Compact handles compaction during a run
> ObservationPack reshapes observation handling
> Evidence-Preserving Reducer covers delegated reading
On the 51-task EdgeBench evaluation, the savings translate to about a third off API cost. In dollars that is an estimated $8.75 to $13.50 per hour against native Codex and Claude Code harnesses, and $4.36 to $5.71 against the baseline harness.
Because the search runs across many environments rather than one, the retained mechanisms keep working outside the setting that produced them. Code is on GitHub under NVlabs.
Paper: https://arxiv.org/abs/2609.20519
Chat with Paper: https://academy.dair.ai/papers/sol-pi-recursively-scaling-auto-research-loops-for-efficient-agent-harness-2609.20519