{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-09-08T18:20:44.688Z","headline":"François Chollet 评 GPT-6 Astra 在 ARC-AGI-3 上的表现","description":"François Chollet 发文称 GPT-6 Astra 在交互式推理任务上带来阶跃式能力提升，使用标准 harness 在 ARC-AGI-3 上得 66%，配合持续对话 harness 和自定义 compaction 接近 100%，每局成本约 $360。","url":"https://www.aioga.com/news/cmtlyeuas0rigrow5d2jq4xb7/","mainEntityOfPage":"https://www.aioga.com/news/cmtlyeuas0rigrow5d2jq4xb7/","datePublished":"2026-09-03T19:42:59.000Z","dateModified":"2026-09-03T19:42:59.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://x.com/fchollet/status/2095598451115614371","https://aihot.news/items/cmtlyeuas0rigrow5d2jq4xb7"],"canonicalUrl":"https://www.aioga.com/news/cmtlyeuas0rigrow5d2jq4xb7/","directAnswer":{"@type":"Answer","text":"François Chollet 发文称，GPT-6 Astra 在交互式推理任务上实现阶跃式能力提升，使用标准 harness 在 ARC-AGI-3 上得分为 66%。","url":"https://www.aioga.com/news/cmtlyeuas0rigrow5d2jq4xb7/","dateCreated":"2026-09-03T19:42:59.000Z","author":{"@type":"Organization","@id":"https://www.aioga.com/authors/aioga-editorial/#editorial-team","name":"Aioga Editorial Team","url":"https://www.aioga.com/authors/aioga-editorial/"}},"evidence":[{"@type":"CreativeWork","name":"@fchollet) source article","url":"https://x.com/fchollet/status/2095598451115614371","datePublished":"2026-09-03T19:42:59.000Z","provider":{"@type":"Organization","name":"@fchollet)","url":"https://x.com/fchollet/status/2095598451115614371"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.news/items/cmtlyeuas0rigrow5d2jq4xb7","datePublished":"2026-09-03T19:42:59.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.news/items/cmtlyeuas0rigrow5d2jq4xb7"}}],"aggregationSource":"@fchollet)","originalPublisher":{"name":"@fchollet)","url":"https://x.com/fchollet/status/2095598451115614371"},"geoDeepAnswer":null,"article":{"id":"cmtlyeuas0rigrow5d2jq4xb7","slug":"cmtlyeuas0rigrow5d2jq4xb7","url":"https://www.aioga.com/news/cmtlyeuas0rigrow5d2jq4xb7/","title":"François Chollet 评 GPT-6 Astra 在 ARC-AGI-3 上的表现","title_en":"","summary":"François Chollet 发文称 GPT-6 Astra 在交互式推理任务上带来阶跃式能力提升，使用标准 harness 在 ARC-AGI-3 上得 66%，配合持续对话 harness 和自定义 compaction 接近 100%，每局成本约 $360。","source":"@fchollet)","sourceUrl":"https://x.com/fchollet/status/2095598451115614371","aiHotUrl":"https://aihot.news/items/cmtlyeuas0rigrow5d2jq4xb7","publishedAt":"2026-09-03T19:42:59.000Z","category":"行业动态","score":72,"selected":true,"articleBody":["François Chollet 发文称 GPT-6 Astra 在交互式推理任务上带来阶跃式能力提升，使用标准 harness 在 ARC-AGI-3 上得 66%，配合持续对话 harness","和自定义 compaction 接近 100%，每局成本约 $360。"],"articleImages":[],"mediaStatus":"none","articleBodyZh":["François Chollet 评 GPT-6 Astra 在 ARC-AGI-3 上的表现 这条更新来自 x.com，发布时间为 2026-09-03，Aioga 保留原文入口以便核验。","摘要：François Chollet 发文称 GPT-6 Astra 在交互式推理任务上带来阶跃式能力提升，使用标准 harness 在 ARC-AGI-3 上得 66%，配合持续对话 harness 和自定义 compaction 接近 100%，每局成本约 $360。","背景：来源还称，配合持续对话 harness 和自定义 compaction 后，GPT-6 Astra 在 ARC-AGI-3 上的表现接近 100%，每局成本约为 360 美元。","Aioga 观察：Aioga 判断：材料的重点既包括标准 harness 下的得分，也包括调整运行方式后的表现与成本；解读时应区分两种设置，避免将结果直接等同。","影响与后续：可能影响：该结果可能提高外界对交互式推理测试及 harness 设计的关注，但单条发文不足以证明结果能够普遍复现，也不代表不同设置下的分数可直接比较。 后续观察：需要核对标准 harness、持续对话 harness 与自定义 compaction 的具体配置，并关注 ARC-AGI-3 得分及每局成本是否获得进一步公开验证。"],"translationStatus":"","bodyOrigin":"social-summary","editorial":{"summary":"François Chollet 发文称，GPT-6 Astra 在交互式推理任务上实现阶跃式能力提升，使用标准 harness 在 ARC-AGI-3 上得分为 66%。","background":"来源还称，配合持续对话 harness 和自定义 compaction 后，GPT-6 Astra 在 ARC-AGI-3 上的表现接近 100%，每局成本约为 360 美元。","viewpoint":"Aioga 判断：材料的重点既包括标准 harness 下的得分，也包括调整运行方式后的表现与成本；解读时应区分两种设置，避免将结果直接等同。","implications":"可能影响：该结果可能提高外界对交互式推理测试及 harness 设计的关注，但单条发文不足以证明结果能够普遍复现，也不代表不同设置下的分数可直接比较。","nextStep":"后续观察：需要核对标准 harness、持续对话 harness 与自定义 compaction 的具体配置，并关注 ARC-AGI-3 得分及每局成本是否获得进一步公开验证。","evidenceRefs":["title","summary","articleBody","source"],"status":"published","aiGenerated":true,"autoApproved":true,"generatedBy":"aioga-editorial:gpt-5.6-sol","reviewedBy":"aioga-editorial-review:gpt-5.6-sol","generatedAt":"2026-09-03T20:34:37.504Z","sourceHash":"92f9780eebe07c47","review":{"approved":true,"groundedness":96,"clarity":94,"duplicationRisk":12,"blockingIssues":[],"notes":["候选内容准确区分了标准 harness 下的 66% 得分，以及持续对话 harness 配合自定义 compaction 后接近 100% 的表现。","对可复现性、配置差异和分数可比性的表述属于明确标注的不确定性分析，没有将其冒充来源事实。"]},"validation":{"passed":true,"mode":"ai-auto","revisions":0,"checks":["schema","length","source-attribution","editorial-labels","inference-boundary","low-source-overlap","no-html","independent-ai-review"]}},"tags":["行业动态","@fchollet)"],"translations":{"zh-CN":{"title":"François Chollet 评 GPT-6 Astra 在 ARC-AGI-3 上的表现","summary":"François Chollet 发文称 GPT-6 Astra 在交互式推理任务上带来阶跃式能力提升，使用标准 harness 在 ARC-AGI-3 上得 66%，配合持续对话 harness 和自定义 compaction 接近 100%，每局成本约 $360。","category":"行业动态","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"François Chollet 评 GPT-6 Astra 在 ARC-AGI-3 上的表现 - Aioga AI资讯","description":"François Chollet 发文称 GPT-6 Astra 在交互式推理任务上带来阶跃式能力提升，使用标准 harness 在 ARC-AGI-3 上得 66%，配合持续对话 harness 和自定义 compaction 接近 100%，每局成本约 $360。","url":"https://www.aioga.com/news/cmtlyeuas0rigrow5d2jq4xb7/","articleBody":["François Chollet 评 GPT-6 Astra 在 ARC-AGI-3 上的表现 这条更新来自 x.com，发布时间为 2026-09-03，Aioga 保留原文入口以便核验。","摘要：François Chollet 发文称 GPT-6 Astra 在交互式推理任务上带来阶跃式能力提升，使用标准 harness 在 ARC-AGI-3 上得 66%，配合持续对话 harness 和自定义 compaction 接近 100%，每局成本约 $360。","背景：来源还称，配合持续对话 harness 和自定义 compaction 后，GPT-6 Astra 在 ARC-AGI-3 上的表现接近 100%，每局成本约为 360 美元。","Aioga 观察：Aioga 判断：材料的重点既包括标准 harness 下的得分，也包括调整运行方式后的表现与成本；解读时应区分两种设置，避免将结果直接等同。","影响与后续：可能影响：该结果可能提高外界对交互式推理测试及 harness 设计的关注，但单条发文不足以证明结果能够普遍复现，也不代表不同设置下的分数可直接比较。 后续观察：需要核对标准 harness、持续对话 harness 与自定义 compaction 的具体配置，并关注 ARC-AGI-3 得分及每局成本是否获得进一步公开验证。"]},"en":{"title":"François Chollet evaluates GPT-6 Astra's performance on ARC-AGI-3","summary":"François Chollet wrote that GPT-6 Astra brings a leap in capability on interactive reasoning tasks, achieving 66% on ARC-AGI-3 using the standard harness, nearly 100% with continuous conversation harness and custom compaction, with a cost of about $360 per session.","category":"Industry","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"François Chollet evaluates GPT-6 Astra's performance on ARC-AGI-3 - Aioga AI News","description":"François Chollet wrote that GPT-6 Astra brings a leap in capability on interactive reasoning tasks, achieving 66% on ARC-AGI-3 using the standard harness, nearly 100% with continuo...","url":"https://www.aioga.com/en/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:23:38.046Z"},"ja":{"title":"フランソワ・ショレが ARC-AGI-3 上での GPT-6 Astra のパフォーマンスを評価","summary":"フランソワ・ショレは投稿で、GPT-6 Astra がインタラクティブな推論タスクで飛躍的な能力向上をもたらし、標準ハーネスを使用すると ARC-AGI-3 で 66% を達成し、継続的な対話ハーネスとカスタムコンパクションを組み合わせるとほぼ 100% に達し、1 回あたりのコストは約 360 ドルであると述べた。","category":"業界動向","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"フランソワ・ショレが ARC-AGI-3 上での GPT-6 Astra のパフォーマンスを評価 - Aioga AIニュース","description":"フランソワ・ショレは投稿で、GPT-6 Astra がインタラクティブな推論タスクで飛躍的な能力向上をもたらし、標準ハーネスを使用すると ARC-AGI-3 で 66% を達成し、継続的な対話ハーネスとカスタムコンパクションを組み合わせるとほぼ 100% に達し、1 回あたりのコストは約 360 ドルであると述べた。","url":"https://www.aioga.com/ja/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:23:47.384Z"},"ko":{"title":"프랑수아 숄레, GPT-6 Astra의 ARC-AGI-3 성능 평가","summary":"프랑수아 숄레는 글에서 GPT-6 Astra가 인터랙티브 추론 과제에서 급격한 능력 향상을 가져왔으며, 표준 하니스로 ARC-AGI-3에서 66%를 달성하고, 지속 대화 하니스와 맞춤형 압축을 결합하면 거의 100%에 도달하며, 한 판 비용은 약 $360이라고 밝혔다.","category":"업계 동향","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"프랑수아 숄레, GPT-6 Astra의 ARC-AGI-3 성능 평가 - Aioga AI 뉴스","description":"프랑수아 숄레는 글에서 GPT-6 Astra가 인터랙티브 추론 과제에서 급격한 능력 향상을 가져왔으며, 표준 하니스로 ARC-AGI-3에서 66%를 달성하고, 지속 대화 하니스와 맞춤형 압축을 결합하면 거의 100%에 도달하며, 한 판 비용은 약 $360이라고 밝혔다.","url":"https://www.aioga.com/ko/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:24:35.591Z"},"es":{"title":"François Chollet evalúa el rendimiento de GPT-6 Astra en ARC-AGI-3","summary":"François Chollet escribió que GPT-6 Astra ha traído una mejora escalonada en las tareas de razonamiento interactivo, obteniendo un 66% en ARC-AGI-3 usando el harness estándar, acercándose al 100% con el harness de conversación continua y la compactación personalizada, con un costo por partida de aproximadamente $360.","category":"Industria","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"François Chollet evalúa el rendimiento de GPT-6 Astra en ARC-AGI-3 - Aioga Noticias de IA","description":"François Chollet escribió que GPT-6 Astra ha traído una mejora escalonada en las tareas de razonamiento interactivo, obteniendo un 66% en ARC-AGI-3 usando el harness estándar, acer...","url":"https://www.aioga.com/es/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:24:33.037Z"},"fr":{"title":"Évaluation par François Chollet des performances de GPT-6 Astra sur ARC-AGI-3","summary":"François Chollet a publié un article affirmant que GPT-6 Astra apporte une amélioration significative des capacités dans les tâches de raisonnement interactif, obtenant 66% sur ARC-AGI-3 en utilisant le harnais standard, et atteignant près de 100% avec le harnais de dialogue continu et la compaction personnalisée, chaque partie coûtant environ 360 $.","category":"Industrie","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"Évaluation par François Chollet des performances de GPT-6 Astra sur ARC-AGI-3 - Aioga Actualités IA","description":"François Chollet a publié un article affirmant que GPT-6 Astra apporte une amélioration significative des capacités dans les tâches de raisonnement interactif, obtenant 66% sur ARC...","url":"https://www.aioga.com/fr/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:25:19.746Z"},"de":{"title":"François Chollet bewertet die Leistung von GPT-6 Astra auf ARC-AGI-3","summary":"François Chollet schrieb, dass GPT-6 Astra bei interaktiven Schlussfolgerungsaufgaben einen sprunghaften Leistungsanstieg erzielt, mit dem Standard-Harness 66 % auf ARC-AGI-3 erreicht wird, in Kombination mit einem kontinuierlichen Gesprächs-Harness und angepasster Verdichtung fast 100 %, die Kosten pro Runde betragen etwa 360 $.","category":"行业动态","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"François Chollet bewertet die Leistung von GPT-6 Astra auf ARC-AGI-3 - Aioga KI-News","description":"François Chollet schrieb, dass GPT-6 Astra bei interaktiven Schlussfolgerungsaufgaben einen sprunghaften Leistungsanstieg erzielt, mit dem Standard-Harness 66 % auf ARC-AGI-3 errei...","url":"https://www.aioga.com/de/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:25:21.619Z"},"pt-BR":{"title":"François Chollet avalia o desempenho do GPT-6 Astra no ARC-AGI-3","summary":"François Chollet escreveu que o GPT-6 Astra trouxe um salto significativo de capacidade em tarefas de raciocínio interativo, alcançando 66% no ARC-AGI-3 usando a harness padrão, e perto de 100% quando combinado com harness de diálogo contínuo e compactação personalizada, com custo por jogo de cerca de $360.","category":"行业动态","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"François Chollet avalia o desempenho do GPT-6 Astra no ARC-AGI-3 - Aioga Notícias de IA","description":"François Chollet escreveu que o GPT-6 Astra trouxe um salto significativo de capacidade em tarefas de raciocínio interativo, alcançando 66% no ARC-AGI-3 usando a harness padrão, e...","url":"https://www.aioga.com/pt-BR/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:26:03.193Z"},"ru":{"title":"Франсуа Шолет оценивает производительность GPT-6 Astra на ARC-AGI-3","summary":"Франсуа Шоле написал, что GPT-6 Astra обеспечивает скачкообразное улучшение возможностей в задачах интерактивного рассуждения, используя стандартный harness набирает 66% на ARC-AGI-3, в сочетании с постоянным диалоговым harness и кастомным compaction достигает почти 100%, стоимость каждой сессии около $360.","category":"行业动态","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"Франсуа Шолет оценивает производительность GPT-6 Astra на ARC-AGI-3 - Aioga Новости ИИ","description":"Франсуа Шоле написал, что GPT-6 Astra обеспечивает скачкообразное улучшение возможностей в задачах интерактивного рассуждения, используя стандартный harness набирает 66% на ARC-AGI...","url":"https://www.aioga.com/ru/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:26:10.513Z"},"ar":{"title":"فرانسوا شولي يقيم أداء GPT-6 Astra على ARC-AGI-3","summary":"كتب فرانسوا شولي أن GPT-6 Astra يجلب قفزة كبيرة في القدرات في مهام الاستدلال التفاعلي، باستخدام harness القياسي حصل على 66% في ARC-AGI-3، ومع الاستفادة من harness للحوار المستمر و compaction المخصص يقترب من 100%، وتكلفة كل جلسة حوالي 360 دولارًا.","category":"行业动态","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"فرانسوا شولي يقيم أداء GPT-6 Astra على ARC-AGI-3 - Aioga أخبار الذكاء الاصطناعي","description":"كتب فرانسوا شولي أن GPT-6 Astra يجلب قفزة كبيرة في القدرات في مهام الاستدلال التفاعلي، باستخدام harness القياسي حصل على 66% في ARC-AGI-3، ومع الاستفادة من harness للحوار المستمر و...","url":"https://www.aioga.com/ar/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:27:09.213Z"},"hi":{"title":"फ़्राँक्वा शोलेट ने ARC-AGI-3 पर GPT-6 Astra के प्रदर्शन का मूल्यांकन किया","summary":"फ्राँस्वा शॉलेट ने लिखा कि GPT-6 Astra ने अंतःक्रियात्मक तर्क कार्यों में कदम-जैसे क्षमता सुधार लाए हैं, स्टैंडर्ड हार्नेस का उपयोग करके ARC-AGI-3 पर 66% स्कोर किया, लगातार संवाद हार्नेस और कस्टम कम्पैक्शन के साथ लगभग 100% तक पहुंचा, प्रत्येक मैच की लागत लगभग $360 है।","category":"行业动态","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"फ़्राँक्वा शोलेट ने ARC-AGI-3 पर GPT-6 Astra के प्रदर्शन का मूल्यांकन किया - Aioga AI समाचार","description":"फ्राँस्वा शॉलेट ने लिखा कि GPT-6 Astra ने अंतःक्रियात्मक तर्क कार्यों में कदम-जैसे क्षमता सुधार लाए हैं, स्टैंडर्ड हार्नेस का उपयोग करके ARC-AGI-3 पर 66% स्कोर किया, लगातार संवाद ह...","url":"https://www.aioga.com/hi/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:27:02.656Z"},"it":{"title":"François Chollet valuta le prestazioni di GPT-6 Astra su ARC-AGI-3","summary":"François Chollet ha scritto che GPT-6 Astra porta un salto nelle capacità nei compiti di ragionamento interattivo, ottenendo il 66% su ARC-AGI-3 usando il harness standard, e quasi il 100% con harness di dialogo continuo e compattazione personalizzata, con un costo per sessione di circa 360 dollari.","category":"行业动态","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"François Chollet valuta le prestazioni di GPT-6 Astra su ARC-AGI-3 - Aioga Notizie IA","description":"François Chollet ha scritto che GPT-6 Astra porta un salto nelle capacità nei compiti di ragionamento interattivo, ottenendo il 66% su ARC-AGI-3 usando il harness standard, e quasi...","url":"https://www.aioga.com/it/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:28:09.046Z"},"nl":{"title":"François Chollet evalueert de prestaties van GPT-6 Astra op ARC-AGI-3","summary":"François Chollet schreef dat GPT-6 Astra een sprongsgewijze verbetering in interactieve redeneertaken biedt, met een score van 66% op ARC-AGI-3 met de standaard harness, bijna 100% in combinatie met de continue gespreks-harness en aangepaste compaction, met een kosten per sessie van ongeveer $360.","category":"行业动态","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"François Chollet evalueert de prestaties van GPT-6 Astra op ARC-AGI-3 - Aioga AI-nieuws","description":"François Chollet schreef dat GPT-6 Astra een sprongsgewijze verbetering in interactieve redeneertaken biedt, met een score van 66% op ARC-AGI-3 met de standaard harness, bijna 100%...","url":"https://www.aioga.com/nl/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:27:58.202Z"},"tr":{"title":"François Chollet, GPT-6 Astra'nın ARC-AGI-3 üzerindeki performansını değerlendiriyor","summary":"François Chollet, GPT-6 Astra'nın etkileşimli akıl yürütme görevlerinde kademeli bir yetenek artışı sağladığını belirtti, standart harness kullanarak ARC-AGI-3'te %66 elde etti, sürekli diyalog harness ve özel sıkıştırma ile %100'e yaklaştı, her oyun başına maliyet yaklaşık 360$.","category":"行业动态","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"François Chollet, GPT-6 Astra'nın ARC-AGI-3 üzerindeki performansını değerlendiriyor - Aioga AI Haberleri","description":"François Chollet, GPT-6 Astra'nın etkileşimli akıl yürütme görevlerinde kademeli bir yetenek artışı sağladığını belirtti, standart harness kullanarak ARC-AGI-3'te %66 elde etti, sü...","url":"https://www.aioga.com/tr/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:28:56.445Z"},"vi":{"title":"François Chollet đánh giá hiệu suất của GPT-6 Astra trên ARC-AGI-3","summary":"François Chollet đăng bài cho biết GPT-6 Astra mang lại cải thiện khả năng đột phá trong các nhiệm vụ suy luận tương tác, sử dụng harness tiêu chuẩn đạt 66% trên ARC-AGI-3, kết hợp với harness đối thoại liên tục và compaction tùy chỉnh gần như 100%, chi phí mỗi trận khoảng 360 đô la.","category":"行业动态","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"François Chollet đánh giá hiệu suất của GPT-6 Astra trên ARC-AGI-3 - Tin tức AI Aioga","description":"François Chollet đăng bài cho biết GPT-6 Astra mang lại cải thiện khả năng đột phá trong các nhiệm vụ suy luận tương tác, sử dụng harness tiêu chuẩn đạt 66% trên ARC-AGI-3, kết hợp...","url":"https://www.aioga.com/vi/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:28:59.780Z"},"id":{"title":"François Chollet menilai kinerja GPT-6 Astra di ARC-AGI-3","summary":"François Chollet menulis bahwa GPT-6 Astra memberikan peningkatan kemampuan yang signifikan dalam tugas penalaran interaktif, menggunakan harness standar di ARC-AGI-3 mendapatkan 66%, dengan harness percakapan berkelanjutan dan compactasi kustom mendekati 100%, biaya per sesi sekitar $360.","category":"行业动态","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"François Chollet menilai kinerja GPT-6 Astra di ARC-AGI-3 - Berita AI Aioga","description":"François Chollet menulis bahwa GPT-6 Astra memberikan peningkatan kemampuan yang signifikan dalam tugas penalaran interaktif, menggunakan harness standar di ARC-AGI-3 mendapatkan 6...","url":"https://www.aioga.com/id/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:29:43.145Z"},"th":{"title":"ฟรองซัว โชเลต์ ประเมินผลการทำงานของ GPT-6 Astra บน ARC-AGI-3","summary":"François Chollet เขียนว่า GPT-6 Astra มีการยกระดับความสามารถในงานการให้เหตุผลแบบโต้ตอบอย่างก้าวกระโดด โดยใช้ standard harness ทำได้ 66% ใน ARC-AGI-3 ร่วมกับการสนทนาต่อเนื่อง harness และการบีบอัดที่กำหนดเอง ใกล้ 100% ต้นทุนต่อรอบประมาณ $360","category":"行业动态","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"ฟรองซัว โชเลต์ ประเมินผลการทำงานของ GPT-6 Astra บน ARC-AGI-3 - ข่าว AI Aioga","description":"François Chollet เขียนว่า GPT-6 Astra มีการยกระดับความสามารถในงานการให้เหตุผลแบบโต้ตอบอย่างก้าวกระโดด โดยใช้ standard harness ทำได้ 66% ใน ARC-AGI-3 ร่วมกับการสนทนาต่อเนื่อง harnes...","url":"https://www.aioga.com/th/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:29:51.888Z"},"pl":{"title":"François Chollet ocenia wyniki GPT-6 Astra w ARC-AGI-3","summary":"François Chollet napisał, że GPT-6 Astra przynosi skokowy wzrost zdolności w zadaniach związanych z interaktywnym rozumowaniem, osiągając 66% przy użyciu standardowego harness na ARC-AGI-3, a w połączeniu z ciągłą rozmową harness i niestandardową kompaktacją zbliża się do 100%, koszt każdej rundy wynosi około 360 dolarów.","category":"行业动态","source":"@fchollet)","aggregationSource":"@fchollet)","pageTitle":"François Chollet ocenia wyniki GPT-6 Astra w ARC-AGI-3 - Aioga Wiadomości AI","description":"François Chollet napisał, że GPT-6 Astra przynosi skokowy wzrost zdolności w zadaniach związanych z interaktywnym rozumowaniem, osiągając 66% przy użyciu standardowego harness na A...","url":"https://www.aioga.com/pl/news/cmtlyeuas0rigrow5d2jq4xb7/","contentTranslated":true,"sourceHash":"6438e8fdaf33394d","translatedAt":"2026-09-03T20:30:38.194Z"}},"evidenceTier":"social-signal","reviewStatus":"editorial-selected","indexable":true,"editorialCover":"/page-visuals/topic-timeline.png"}}