ARC Prize’s ARC-AGI-3 showdown landed with a split verdict. In the official setup, Anthropic’s Opus 5 scored 30.2%, while OpenAI’s GPT-5.6 Sol posted 7.8%, based on the published results. OpenAI says that gap looks very different when the model gets to carry its reasoning across turns: in the company’s own rerun, GPT-5.6 Sol reached 38.3%.
That claim matters because ARC-AGI-3 is meant to test what its creators call fluid intelligence through novel, game-like tasks. These aren’t recall questions. They’re supposed to show how well a model reasons through something new. OpenAI’s argument is that the official harness undersells GPT-5.6 Sol because it throws away reasoning between turns. In OpenAI’s rerun through the Responses API, Retained Reasoning and Compaction pushed the score to 38.3% by keeping continuity intact and compressing older context into summaries instead of just cutting it off.
ARC Prize describes the benchmark as provider-neutral. Still, reporting says the official harness may have used an outdated API for GPT-5.6 Sol. François Chollet has also reportedly said provider-specific settings can be acceptable when they’re general-purpose and clearly disclosed. So the disagreement hasn’t really gone away. The open question is whether OpenAI’s rerun belongs in the same comparison, or whether it should be treated as a separate result.
For anyone sizing up models for real deployments, that’s the part to keep in view. OpenAI’s 38.3% is still self-reported, and it’s on the public ARC-AGI-3 set. Independent replication on the private dataset still hasn’t happened. And for now, the official leaderboard still has Anthropic in front.
Author: Anthony John Padilla
{
"social": {
"email": "content.reviewer90@ext.softonic.com",
"facebook": "",
"twitter": "",
"linkedin": ""
},
"ja-JP": "",
"de-DE": "Anthony Padilla ist ein auf den Philippinen ansässiger Schriftsteller mit Erfahrung in redaktionellen, digitalen und markenorientierten Inhalten. Seine Arbeit konzentriert sich darauf, komplexe Ideen in klare, verständliche Texte umzuwandeln.",
"en-US": "Anthony Padilla is a Philippines-based writer with experience across editorial, digital, and brand-focused content. His work centers on translating complex ideas into clear, accessible copy, with a strong emphasis on structure, accuracy, and user-focused communication. He has contributed to long-form articles, product and platform copy, and marketing-led editorial pieces.",
"es-ES": "Anthony Padilla es un escritor con base en Filipinas con experiencia en contenido editorial, digital y enfocado en marca. Su trabajo se centra en traducir ideas complejas en textos claros y accesibles.",
"fr-FR": "Anthony Padilla est un écrivain basé aux Philippines ayant de l'expérience dans le contenu éditorial, numérique et axé sur la marque. Son travail consiste à traduire des idées complexes en textes clairs et accessibles.",
"it-IT": "Anthony Padilla è uno scrittore con base nelle Filippine con esperienza in contenuti editoriali, digitali e focalizzati sul brand. Il suo lavoro si concentra sulla traduzione di idee complesse in testi chiari e accessibili.",
"nl-NL": "Anthony Padilla is een schrijver gevestigd op de Filipijnen met ervaring in redactionele, digitale en merkgerichte content. Zijn werk richt zich op het vertalen van complexe ideeën naar helder, begrijpelijke teksten.",
"pl-PL": "Anthony Padilla to pisarz mieszkający na Filipinach z doświadczeniem w tworzeniu treści redakcyjnych, cyfrowych i skoncentrowanych na marce. Jego prace skupiają się na przekształcaniu skomplikowanych idei w jasne, zrozumiałe teksty.",
"pt-BR": "Anthony Padilla é um escritor baseado nas Filipinas com experiência em conteúdo editorial, digital e focado em marca. Seu trabalho se concentra em traduzir ideias complexas em textos claros e acessíveis."
}
View all posts by Anthony John Padilla
