Former OpenAI researcher launches Context Engine: after agents attach memory, they achieve low-level reasoning parity with human reasoning

robot
Abstract generation in progress

Crypto界消息,前OpenAI研究员创立的Applied Compute发布了Context Engine,这是一套企业代理的上下文引擎。该引擎将内部文档、工单历史和代理运行轨迹提炼成知识库contextbase,代理在执行任务时可以直接检索,降低推理成本。在apex-agents(Mercor开发的投行、咨询、法律专业评测)上,GPT-5.4的低推理档从44.5%提升至52.4%,与不挂记忆的中推理档52.3%基本持平。低推理提升7.9%,中推理提升3.7%,极高推理反而下降0.7%。在中推理基线上,apex-agents的GPT-5.4从44.2%提升至51.7%,相对提升16.9%;GPT-5.4-mini从33.4%提升至38.7%,相对提升15.8%。Applied Compute认为,任务间可复用结构少,基线分数已接近天花板。

View Original
This page may contain third-party content, which is provided for information purposes only (not representations/warranties) and should not be considered as an endorsement of its views by Gate, nor as financial or professional advice. See Disclaimer for details.
  • Reward
  • Comment
  • Repost
  • Share
Comment
Add a comment
Add a comment
No comments
  • Pin