# In harmony with gpt-oss

arXiv: https://arxiv.org/abs/2604.00362 · HTML: https://mlsanity.com/paper/2604.00362/

## Key facts

- Cited by 1 papers in the map (data through 14 Oct 2026).
- Builds on 2 of its 12 references that are in the map.
- First posted on arXiv 1 Apr 2026.

## Builds on (most cited first, 2 of 2)

- [SWE-agent: Agent-Computer Interfaces Enable Automated Software Engineering](https://mlsanity.com/paper/2405.15793/) (2024)
- [Is GPT-OSS Good? A Comprehensive Evaluation of OpenAI's Latest Open Source Models](https://mlsanity.com/paper/2508.12461/) (2025)

## Built on by (most cited first, 1 of 1)

- [Does the Same Token Mean the Same State? MoE Routing as Signal for Reasoning Control](https://mlsanity.com/paper/2606.22798/) (2026)

## Abstract (arXiv)

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%).