--- title: Nvidia just showed that the harness, not the AI model,... description: Nvidia research published on the ARC-AGI-3 interactive reasoning benchmark shows that Claude Opus 5, when run through a custom harness with memory handling and... canonical: https://daily.dev/posts/nvidia-just-showed-that-the-harness-not-the-ai-model-is-now-the-real-hero-xhs3liky3 twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Nvidia just showed that the harness, not the AI model, is now the real hero | daily.dev og:description: Nvidia research published on the ARC-AGI-3 interactive reasoning benchmark shows that Claude Opus 5, when run through a custom harness with memory handling and... og:url: https://daily.dev/posts/nvidia-just-showed-that-the-harness-not-the-ai-model-is-now-the-real-hero-xhs3liky3 og:image: https://api.daily.dev/og/posts/xHs3LIKy3.png og:image:alt: Nvidia just showed that the harness, not the AI model, is now the real hero og:image:width: 1200 og:image:height: 630 og:locale: en --- > ## Documentation Index > Fetch the complete documentation index at: https://daily.dev/llms.txt > Use this file to discover all available pages before exploring further. # Nvidia just showed that the harness, not the AI model, is now the real hero **[TechCrunch](https://daily.dev/sources/tc)** · 5 min read · 0 upvotes · 0 comments ## Summary Nvidia research published on the ARC-AGI-3 interactive reasoning benchmark shows that Claude Opus 5, when run through a custom harness with memory handling and a supervisor agent, achieved a 100% score, versus only 30% without the harness (the best raw-model score among all tested). This is evidence that the scaffolding around a model, called the harness, matters more for long-horizon agentic tasks than the underlying model itself. OpenAI's own harness tweaks tripled its scores but still fell far short of 100%. Databricks research is cited separately showing harness choice can also double inference costs regardless of model choice. Nvidia frames this as an argument for open harnesses (its Nemo ecosystem) giving users more control than open or closed models alone. ## Full article daily.dev links to this article rather than hosting it. Read it at the original source: ## Questions this post answers ### What score did Claude Opus 5 get on ARC-AGI-3 with and without a custom agent harness? Claude Opus 5 scored 100% on the ARC-AGI-3 interactive reasoning benchmark when run through Nvidia's custom Agentic Variation Operators (AVO) harness, which included memory handling and a supervisor component. Without the harness, the same model scored only 30%, which was still the best result among all models tested without a custom harness. _Anyone weighing model versus harness choices for agentic tasks tracks benchmarks like this on daily.dev._ ### How much did OpenAI improve its ARC-AGI-3 scores by tweaking its agent harness settings? OpenAI tripled its models' ARC-AGI-3 scores simply by tweaking two settings in the harness, though the results still stayed under 10% before tweaking and never approached the 100% Nvidia achieved with a supervisor-equipped harness. This came after OpenAI's models initially scored poorly, prompting its own internal research into harness tuning. _Developers comparing agent harness strategies across labs follow these results on daily.dev._ ### Can the choice of agent harness affect AI inference costs even when using the same model? Yes, Databricks research found that using the wrong harness with the same underlying model can double inference costs. Databricks CEO Ali Ghodsi noted that cost differences often blamed on model choice are actually caused by harness inefficiency, since the same model run through different harnesses produces significantly different costs. _Teams budgeting for agentic workloads track harness-versus-model cost findings on daily.dev._ --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#nvidia](https://daily.dev/tags/nvidia), [#claude](https://daily.dev/tags/claude) [View this post on daily.dev](https://daily.dev/posts/nvidia-just-showed-that-the-harness-not-the-ai-model-is-now-the-real-hero-xhs3liky3) ```json {"@context":"https://schema.org","@graph":[{"@type":"Organization","@id":"https://daily.dev/#organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180},"sameAs":["https://twitter.com/dailydotdev","https://github.com/dailydotdev","https://www.linkedin.com/company/daily-dev-ltd"]},{"@type":"WebSite","@id":"https://daily.dev/#website","url":"https://daily.dev","name":"daily.dev","publisher":{"@id":"https://daily.dev/#organization"},"potentialAction":{"@type":"SearchAction","target":{"@type":"EntryPoint","urlTemplate":"https://daily.dev/search?q={search_term_string}"},"query-input":"required name=search_term_string"}}]} {"@context":"https://schema.org","@type":"TechArticle","headline":"Nvidia just showed that the harness, not the AI model, is now the real hero","url":"https://daily.dev/posts/nvidia-just-showed-that-the-harness-not-the-ai-model-is-now-the-real-hero-xhs3liky3","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/nvidia-just-showed-that-the-harness-not-the-ai-model-is-now-the-real-hero-xhs3liky3"},"datePublished":"2026-08-21T19:45:52.957Z","dateModified":"2026-08-22T21:19:16.876Z","description":"Nvidia research published on the ARC-AGI-3 interactive reasoning benchmark shows that Claude Opus 5, when run through a custom harness with memory handling and...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/403b98c99dd54231226eb85589147594?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/403b98c99dd54231226eb85589147594?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"TechCrunch","inLanguage":"en","publisher":{"@type":"Organization","name":"daily.dev","url":"https://daily.dev","logo":{"@type":"ImageObject","url":"https://daily.dev/apple-touch-icon.png","width":180,"height":180}},"author":{"@type":"Organization","name":"TechCrunch","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/tc","url":"https://daily.dev/sources/tc"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/nvidia-just-showed-that-the-harness-not-the-ai-model-is-now-the-real-hero-xhs3liky3","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":0},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,nvidia,claude","timeRequired":"PT5M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"TechCrunch","item":"https://daily.dev/sources/tc"},{"@type":"ListItem","position":3,"name":"Nvidia just showed that the harness, not the AI model, is now the real hero"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/nvidia-just-showed-that-the-harness-not-the-ai-model-is-now-the-real-hero-xhs3liky3#faq","mainEntity":[{"@type":"Question","name":"What score did Claude Opus 5 get on ARC-AGI-3 with and without a custom agent harness?","acceptedAnswer":{"@type":"Answer","text":"Claude Opus 5 scored 100% on the ARC-AGI-3 interactive reasoning benchmark when run through Nvidia's custom Agentic Variation Operators (AVO) harness, which included memory handling and a supervisor component. Without the harness, the same model scored only 30%, which was still the best result among all models tested without a custom harness. Anyone weighing model versus harness choices for agentic tasks tracks benchmarks like this on daily.dev."}},{"@type":"Question","name":"How much did OpenAI improve its ARC-AGI-3 scores by tweaking its agent harness settings?","acceptedAnswer":{"@type":"Answer","text":"OpenAI tripled its models' ARC-AGI-3 scores simply by tweaking two settings in the harness, though the results still stayed under 10% before tweaking and never approached the 100% Nvidia achieved with a supervisor-equipped harness. This came after OpenAI's models initially scored poorly, prompting its own internal research into harness tuning. Developers comparing agent harness strategies across labs follow these results on daily.dev."}},{"@type":"Question","name":"Can the choice of agent harness affect AI inference costs even when using the same model?","acceptedAnswer":{"@type":"Answer","text":"Yes, Databricks research found that using the wrong harness with the same underlying model can double inference costs. Databricks CEO Ali Ghodsi noted that cost differences often blamed on model choice are actually caused by harness inefficiency, since the same model run through different harnesses produces significantly different costs. Teams budgeting for agentic workloads track harness-versus-model cost findings on daily.dev."}}]} ```