复现GPT-OSS-20B的基准成绩，突破官方未公开工具的限制。
In harmony with gpt-oss
- 逆向解析模型内建工具，发现其具备高置信度的工具调用先验。
- 构建原生消息格式代理框架，避免接口转换损失信息。
- 首次独立复现官方公布的60.4%、53.3%和91.7%关键指标。
目前无人能使用工具独立复现OpenAI发布的GPT-OSS-20B得分，因原始论文既未披露使用的工具，也未公开代理架构。本文通过逆向工程发现，即使在未提供工具定义的情况下，GPT-OSS仍能以高统计置信度调用训练分布中的工具——这反映的是强大的先验知识，而非幻觉。我们构建了原生harmony代理框架（https://github.com/borislavmavrin/harmonyagent.git），以模型原生格式编码消息，绕过有损的Chat Completions转换。由此实现首个独立复现：在SWE Verified HIGH上达到60.4%（官方公布60.7%），MEDIUM为53.3%（53.2%），使用工具时在AIME25上达91.7%（90.4%）。
原文摘要 · Abstract (English)
No one has independently reproduced OpenAI's published scores for gpt-oss-20b with tools, because the original paper discloses neither the tools nor the agent harness. We reverse-engineered the model's in-distribution tools: when prompted without tool definitions, gpt-oss still calls tools from its training distribution with high statistical confidence -- a strong prior, not a hallucination. We then built a native harmony agent harness (https://github.com/borislavmavrin/harmonyagent.git) that encodes messages in the model's native format, bypassing the lossy Chat Completions conversion. Together, these yield the first independent reproduction of OpenAI's published scores: 60.4% on SWE Verified HIGH (published 60.7%), 53.3% MEDIUM (53.2%), and 91.7% on AIME25 with tools (90.4%).
Thank you to arXiv for use of its open access interoperability. PaperDance 不是 arXiv 官方产品；中文卡片由大模型生成，请以原文为准。