--- title: NVIDIA AVO Reaches 100% on ARC-AGI-3, Demonstrating a... description: NVIDIA's AVO (Agentic Variation Operators) agent architecture achieved a perfect 100.00 RHAE score on the ARC-AGI-3 public benchmark, completing all 183 levels... canonical: https://daily.dev/posts/nvidia-avo-reaches-100-on-arc-agi-3-demonstrating-a-frontier-level-general-purpose-architecture-fo-mwxfl3mwk twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: NVIDIA AVO Reaches 100% on ARC-AGI-3, Demonstrating a Frontier-Level General-Purpose Architecture for Long-Horizon Autonomous Agents | daily.dev og:description: NVIDIA's AVO (Agentic Variation Operators) agent architecture achieved a perfect 100.00 RHAE score on the ARC-AGI-3 public benchmark, completing all 183 levels... og:url: https://daily.dev/posts/nvidia-avo-reaches-100-on-arc-agi-3-demonstrating-a-frontier-level-general-purpose-architecture-fo-mwxfl3mwk og:image: https://api.daily.dev/og/posts/mwxfl3mwK.png og:image:alt: NVIDIA AVO Reaches 100% on ARC-AGI-3, Demonstrating a Frontier-Level General-Purpose Architecture for Long-Horizon Autonomous Agents 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 AVO Reaches 100% on ARC-AGI-3, Demonstrating a Frontier-Level General-Purpose Architecture for Long-Horizon Autonomous Agents **[NVIDIA Developer](https://daily.dev/sources/nvidiadev)** · 9 min read · 1 upvotes · 0 comments ## Summary NVIDIA's AVO (Agentic Variation Operators) agent architecture achieved a perfect 100.00 RHAE score on the ARC-AGI-3 public benchmark, completing all 183 levels across 25 environments using 6,624 environment actions with Claude Opus 5, about 12% fewer actions than the comparable VISTA system. AVO was originally developed for autonomous GPU-kernel optimization, where it ran continuously for seven days, explored 500+ optimization directions, and produced attention kernels outperforming cuDNN by up to 3.5% and FlashAttention-4 by up to 10.5% on DGX B200 hardware. The same general-purpose agent architecture, featuring persistent memory and a supervisor mechanism, transferred from software/kernel engineering to an unfamiliar interactive reasoning task without domain-specific redesign, suggesting that long-horizon agent capability depends more on system-level machinery than on the underlying model 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 NVIDIA's AVO agent achieve on the ARC-AGI-3 public benchmark? NVIDIA's AVO agent architecture achieved a perfect 100.00 RHAE score across all 25 environments of the ARC-AGI-3 public set, completing all 183 levels. Using Claude Opus 5 as the underlying model, it did this in 6,624 environment actions, about 12% fewer than VISTA's reported 7,542 actions for the same levels. _See daily.dev for ongoing coverage of how agent architectures are benchmarked against models like ARC-AGI-3._ ### How much faster were AVO's evolved attention kernels compared to cuDNN and FlashAttention-4? AVO-evolved multihead attention kernels outperformed cuDNN by up to 3.5% and FlashAttention-4 by up to 10.5% across evaluated configurations on NVIDIA DGX B200 systems. The agent ran autonomously for seven days, explored over 500 optimization directions, and produced 40 committed kernel versions before adapting the result to grouped-query attention in about 30 minutes. _Engineers tuning GPU kernels can follow benchmark comparisons like this one on daily.dev._ ### What is the difference between AVO and VISTA as agent architectures for ARC-AGI-3? AVO is NVIDIA's own long-horizon agent architecture with persistent memory, a supervisory monitoring layer, and its own execution loop, whereas VISTA instantiates its harness through Claude Code or Codex using Claude Opus 5 or GPT-5.6 Sol directly. Both used direct-interaction task design, but AVO's text-only 64x64 grid observation and memory system are distinct implementation choices from VISTA's rendered PNG approach. _Teams comparing agent harness designs can track architecture trade-offs like these on daily.dev._ ## Similar posts on daily.dev - [NVIDIA Kaggle Grandmasters Win Artificial General Intelligence Competition](https://daily.dev/posts/nvidia-kaggle-grandmasters-win-artificial-general-intelligence-competition-nkdcmanze) · NVIDIA Developer · 1 upvotes · 0 comments - [What Actually Mattered in AI — August 2026](https://daily.dev/posts/what-actually-mattered-in-ai-august-2026-ybkm1vfpo) · Medium · 1 upvotes · 0 comments - [Leaderboard](https://daily.dev/posts/leaderboard-bzhhktbcq) · Hacker News · 1 upvotes · 0 comments --- Tags: [#ai-agents](https://daily.dev/tags/ai-agents), [#nvidia](https://daily.dev/tags/nvidia), [#gpu](https://daily.dev/tags/gpu), [#claude](https://daily.dev/tags/claude) [View this post on daily.dev](https://daily.dev/posts/nvidia-avo-reaches-100-on-arc-agi-3-demonstrating-a-frontier-level-general-purpose-architecture-fo-mwxfl3mwk) ```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 AVO Reaches 100% on ARC-AGI-3, Demonstrating a Frontier-Level General-Purpose Architecture for Long-Horizon Autonomous Agents","url":"https://daily.dev/posts/nvidia-avo-reaches-100-on-arc-agi-3-demonstrating-a-frontier-level-general-purpose-architecture-fo-mwxfl3mwk","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/nvidia-avo-reaches-100-on-arc-agi-3-demonstrating-a-frontier-level-general-purpose-architecture-fo-mwxfl3mwk"},"datePublished":"2026-08-21T13:02:05.613Z","dateModified":"2026-10-15T02:22:53.587Z","description":"NVIDIA's AVO (Agentic Variation Operators) agent architecture achieved a perfect 100.00 RHAE score on the ARC-AGI-3 public benchmark, completing all 183 levels...","image":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4346c91c5f6c80470d13951b462cc4ec?_a=AQAEuop","thumbnailUrl":"https://media.daily.dev/image/upload/f_auto,q_auto/v1/posts/4346c91c5f6c80470d13951b462cc4ec?_a=AQAEuop","isAccessibleForFree":true,"articleSection":"NVIDIA Developer","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":"NVIDIA Developer","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/86e45aab42ba48ce83103d01b1119910","url":"https://daily.dev/sources/nvidiadev"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/nvidia-avo-reaches-100-on-arc-agi-3-demonstrating-a-frontier-level-general-purpose-architecture-fo-mwxfl3mwk","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"ai-agents,nvidia,gpu,claude","timeRequired":"PT9M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"NVIDIA Developer","item":"https://daily.dev/sources/nvidiadev"},{"@type":"ListItem","position":3,"name":"NVIDIA AVO Reaches 100% on ARC-AGI-3, Demonstrating a Frontier-Level General-Purpose Architecture for Long-Horizon Autonomous Agents"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/nvidia-avo-reaches-100-on-arc-agi-3-demonstrating-a-frontier-level-general-purpose-architecture-fo-mwxfl3mwk#faq","mainEntity":[{"@type":"Question","name":"What score did NVIDIA's AVO agent achieve on the ARC-AGI-3 public benchmark?","acceptedAnswer":{"@type":"Answer","text":"NVIDIA's AVO agent architecture achieved a perfect 100.00 RHAE score across all 25 environments of the ARC-AGI-3 public set, completing all 183 levels. Using Claude Opus 5 as the underlying model, it did this in 6,624 environment actions, about 12% fewer than VISTA's reported 7,542 actions for the same levels. See daily.dev for ongoing coverage of how agent architectures are benchmarked against models like ARC-AGI-3."}},{"@type":"Question","name":"How much faster were AVO's evolved attention kernels compared to cuDNN and FlashAttention-4?","acceptedAnswer":{"@type":"Answer","text":"AVO-evolved multihead attention kernels outperformed cuDNN by up to 3.5% and FlashAttention-4 by up to 10.5% across evaluated configurations on NVIDIA DGX B200 systems. The agent ran autonomously for seven days, explored over 500 optimization directions, and produced 40 committed kernel versions before adapting the result to grouped-query attention in about 30 minutes. Engineers tuning GPU kernels can follow benchmark comparisons like this one on daily.dev."}},{"@type":"Question","name":"What is the difference between AVO and VISTA as agent architectures for ARC-AGI-3?","acceptedAnswer":{"@type":"Answer","text":"AVO is NVIDIA's own long-horizon agent architecture with persistent memory, a supervisory monitoring layer, and its own execution loop, whereas VISTA instantiates its harness through Claude Code or Codex using Claude Opus 5 or GPT-5.6 Sol directly. Both used direct-interaction task design, but AVO's text-only 64x64 grid observation and memory system are distinct implementation choices from VISTA's rendered PNG approach. Teams comparing agent harness designs can track architecture trade-offs like these on daily.dev."}}]} ```