--- title: Frontier Models with Our Harness Achieve ~99% on... description: Schema is a harness that enables frontier LLMs to achieve ~99% on the ARC-AGI-3 public benchmark by having models act like physicists: they write each game's... canonical: https://daily.dev/posts/frontier-models-with-our-harness-achieve-99-on-arc-agi-3-public-schema-dqa4is0zk twitter:card: summary_large_image twitter:site: @dailydotdev og:type: website og:site_name: daily.dev og:title: Frontier Models with Our Harness Achieve ~99% on ARC-AGI-3 Public — Schema | daily.dev og:description: Schema is a harness that enables frontier LLMs to achieve ~99% on the ARC-AGI-3 public benchmark by having models act like physicists: they write each game's... og:url: https://daily.dev/posts/frontier-models-with-our-harness-achieve-99-on-arc-agi-3-public-schema-dqa4is0zk og:image: https://api.daily.dev/og/posts/DqA4is0Zk.png og:image:alt: Frontier Models with Our Harness Achieve ~99% on ARC-AGI-3 Public — Schema 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. # Frontier Models with Our Harness Achieve ~99% on ARC-AGI-3 Public — Schema **[Hacker News](https://daily.dev/sources/hn)** · 17 min read · 1 upvotes · 0 comments ## Summary Schema is a harness that enables frontier LLMs to achieve ~99% on the ARC-AGI-3 public benchmark by having models act like physicists: they write each game's mechanics as an executable program, verify it against recorded history, and plan inside it using search. The core insight is jointly solving state grounding (what objects and variables exist) and mechanism discovery (how state transitions work) in a single editable program. A controlled comparison shows Schema improves Claude Code baseline from 42.83% to 98.98% RHAE on the 25 public games. The post includes detailed case studies comparing Opus 4.8 and Fable 5 model behaviors, showing that the harness reduces the cost of using a theory while the underlying model determines the cost of discovering one. Results are self-reported and not yet independently verified by ARC Prize. ## 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 Schema achieve on the ARC-AGI-3 public benchmark and how does it compare to official results? Schema self-reports 98.98% Relative Human Action Efficiency (RHAE) on the ARC-AGI-3 public set using an Opus 4.8/Fable 5 pairing, and 95.35% with a Sol xhigh/Sol max pairing. The official verified best from ARC Prize, achieved by Sol max alone, is 13.33% on the public set and 7.78% on the semi-private set. Schema's numbers are self-reported and not independently verified. _Track fast-moving agent benchmark claims like these as they get verified by following AI research on daily.dev._ ### How does the Schema harness improve LLM agent performance over a standard coding agent harness like Claude Code? Using the same Opus 4.8 and Fable 5 model pairing, Schema reaches 98.98% RHAE versus 42.83% under a generic Claude Code harness, a 56.15 percentage point gain from process alone. Schema enforces encoding the world model as a runnable step() program, validating it against every recorded transition before planning, and discarding remaining plans after any prediction error. _Compare agent harness designs like this before choosing one for your own coding-agent workflows on daily.dev._ ### What is Relative Human Action Efficiency (RHAE) in the ARC-AGI-3 benchmark? RHAE compares an agent's per-level action count against a first-exposure human baseline, with the per-level score computed as (human actions divided by agent actions) squared, capped at 1.15. Levels within a game receive increasing weights so later levels count more, and a completion cap prevents a game from scoring 100% unless every level is cleared; the benchmark score averages the game scores. _Understand benchmark metrics like this one when evaluating claims about agentic AI progress on daily.dev._ --- Tags: [#llm](https://daily.dev/tags/llm), [#ai-agents](https://daily.dev/tags/ai-agents), [#world-models](https://daily.dev/tags/world-models) [View this post on daily.dev](https://daily.dev/posts/frontier-models-with-our-harness-achieve-99-on-arc-agi-3-public-schema-dqa4is0zk) ```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":"Frontier Models with Our Harness Achieve ~99% on ARC-AGI-3 Public — Schema","url":"https://daily.dev/posts/frontier-models-with-our-harness-achieve-99-on-arc-agi-3-public-schema-dqa4is0zk","mainEntityOfPage":{"@type":"WebPage","@id":"https://daily.dev/posts/frontier-models-with-our-harness-achieve-99-on-arc-agi-3-public-schema-dqa4is0zk"},"datePublished":"2026-07-16T19:50:12.743Z","dateModified":"2026-09-14T08:01:57.982Z","description":"Schema is a harness that enables frontier LLMs to achieve ~99% on the ARC-AGI-3 public benchmark by having models act like physicists: they write each game's...","image":"https://media.daily.dev/image/upload/s--P4t4XyoV--/f_auto/v1722860399/public/Placeholder%2001","thumbnailUrl":"https://media.daily.dev/image/upload/s--P4t4XyoV--/f_auto/v1722860399/public/Placeholder%2001","isAccessibleForFree":true,"articleSection":"Hacker News","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":"Hacker News","logo":"https://media.daily.dev/image/upload/t_logo,f_auto/v1/logos/hn","url":"https://daily.dev/sources/hn"},"commentCount":0,"discussionUrl":"https://daily.dev/posts/frontier-models-with-our-harness-achieve-99-on-arc-agi-3-public-schema-dqa4is0zk","interactionStatistic":[{"@type":"InteractionCounter","interactionType":{"@type":"LikeAction"},"userInteractionCount":1},{"@type":"InteractionCounter","interactionType":{"@type":"CommentAction"},"userInteractionCount":0}],"keywords":"llm,ai-agents,world-models","timeRequired":"PT17M"} {"@context":"https://schema.org","@type":"BreadcrumbList","itemListElement":[{"@type":"ListItem","position":1,"name":"Home","item":"https://daily.dev"},{"@type":"ListItem","position":2,"name":"Hacker News","item":"https://daily.dev/sources/hn"},{"@type":"ListItem","position":3,"name":"Frontier Models with Our Harness Achieve ~99% on ARC-AGI-3 Public — Schema"}]} {"@context":"https://schema.org","@type":"FAQPage","@id":"https://daily.dev/posts/frontier-models-with-our-harness-achieve-99-on-arc-agi-3-public-schema-dqa4is0zk#faq","mainEntity":[{"@type":"Question","name":"What score did Schema achieve on the ARC-AGI-3 public benchmark and how does it compare to official results?","acceptedAnswer":{"@type":"Answer","text":"Schema self-reports 98.98% Relative Human Action Efficiency (RHAE) on the ARC-AGI-3 public set using an Opus 4.8/Fable 5 pairing, and 95.35% with a Sol xhigh/Sol max pairing. The official verified best from ARC Prize, achieved by Sol max alone, is 13.33% on the public set and 7.78% on the semi-private set. Schema's numbers are self-reported and not independently verified. Track fast-moving agent benchmark claims like these as they get verified by following AI research on daily.dev."}},{"@type":"Question","name":"How does the Schema harness improve LLM agent performance over a standard coding agent harness like Claude Code?","acceptedAnswer":{"@type":"Answer","text":"Using the same Opus 4.8 and Fable 5 model pairing, Schema reaches 98.98% RHAE versus 42.83% under a generic Claude Code harness, a 56.15 percentage point gain from process alone. Schema enforces encoding the world model as a runnable step() program, validating it against every recorded transition before planning, and discarding remaining plans after any prediction error. Compare agent harness designs like this before choosing one for your own coding-agent workflows on daily.dev."}},{"@type":"Question","name":"What is Relative Human Action Efficiency (RHAE) in the ARC-AGI-3 benchmark?","acceptedAnswer":{"@type":"Answer","text":"RHAE compares an agent's per-level action count against a first-exposure human baseline, with the per-level score computed as (human actions divided by agent actions) squared, capped at 1.15. Levels within a game receive increasing weights so later levels count more, and a completion cap prevents a game from scoring 100% unless every level is cleared; the benchmark score averages the game scores. Understand benchmark metrics like this one when evaluating claims about agentic AI progress on daily.dev."}}]} ```