In harmony with gpt-oss

View PDF HTML (experimental)

Abstract: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 (this https URL) 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%).

Subjects:

Artificial Intelligence (cs.AI); Machine Learning (cs.LG)

Cite as: arXiv:2604.00362 [cs.AI]

(or arXiv:2604.00362v1 [cs.AI] for this version)

https://doi.org/10.48550/arXiv.2604.00362

arXiv-issued DOI via DataCite (pending registration)

Submission history

From: Borislav Mavrin [view email] [v1] Wed, 1 Apr 2026 01:16:13 UTC (139 KB)