Intel / article

Six Agent Harness Capabilities for Higher Model Performance

Source: developer.nvidia.com
Author: Michelle Horton
Date: Published Jul 27, 2026

Context

NVIDIA's experimental NOOA framework keeps full tool results in live Python objects and shows the model bounded previews. The model can operate on those objects through code, reducing the need to repeatedly insert their contents into the conversation. NVIDIA's research account reports 82.2% on the coding benchmark with OpenAI's GPT-5.5 at approximately 1.1 million tokens per task, compared with 78.2% and 2.2 million for one comparison . That combination suggests the software surrounding a model can affect both performance and text-processing demand. The reported token reduction is specific to this comparison, not a universal halving of operating costs.

Terms in this piece · Glossary

AI agent — An AI system that doesn't just answer once but works toward a goal in a loop — taking actions, reading the results, and deciding what to do next.
SWE-bench — The standard benchmark for AI coding agents: real GitHub issues from real repositories, scored by whether the agent's patch passes the project's own tests.
agent harness — The scaffolding around a model that turns it into a working agent — the loop, the tools it can call, and the rules for when to stop.
context window — The maximum amount of text a model can consider at once — its working memory for the current conversation or task.

Key quotes

“Harness design alone can account for double-digit swings in benchmark results and significant differences in token cost, with the same underlying model.”
Ricardo Silveira Cabral and Paul Furgale

“NOOA takes a simpler approach: an agent is a single Python class . Its methods are its capabilities. Its fields are its state. Its docstrings are its prompts. Its type annotations are enforced contracts.”
Ricardo Silveira Cabral and Paul Furgale

“On SWE-bench Verified, NOOA reaches 82.2% with GPT-5.5 above the published leaderboard SOTA at submission (79.2%)—and 79.8% with Opus 4.6, using a general-purpose, 253-line agent with no benchmark-specific prompts.”
Ricardo Silveira Cabral and Paul Furgale

“With GPT-5.5, NOOA reaches 82.2% on SWE-bench Verified using 29 LLM calls and ~1.1M tokens per task. The comparison harnesses need 66 calls and 2.2M tokens to reach 78.2%, and 29 calls at 1.3M to reach 78.6%. Parity or better, at roughly half the cost.”
Ricardo Silveira Cabral and Paul Furgale

Read the source developer.nvidia.com