{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-09-08T18:20:44.688Z","headline":"Greg Brockman 转发：GPT-6 Astra 在 ARC-AGI-3 达到 SOTA，基准趋于饱和","description":"Greg Brockman 转发 @arcprize 的评测称 OpenAI 的 GPT-6 Astra 在 ARC-AGI-3 上取得 SOTA，他称该基准已饱和。Astra 标准 harness 得分 63%，经新的 Provider Adapter harness 达 99%，在 96% 的 ARC-AGI-3 关卡上超越人类表现；排行榜图还显示更高推理层级通常成本更低，因为 Astra 用更少动作通关，减少模型调用和 token 数。","url":"https://www.aioga.com/news/cmtm2suk00171rooej14mutgj/","mainEntityOfPage":"https://www.aioga.com/news/cmtm2suk00171rooej14mutgj/","datePublished":"2026-09-03T21:46:00.000Z","dateModified":"2026-09-03T21:46:00.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://x.com/gdb/status/2095629409017614390","https://aihot.news/items/cmtm2suk00171rooej14mutgj"],"canonicalUrl":"https://www.aioga.com/news/cmtm2suk00171rooej14mutgj/","directAnswer":{"@type":"Answer","text":"据 Greg Brockman 转发的 @arcprize 评测，材料称 OpenAI 的 GPT-6 Astra 在 ARC-AGI-3 取得 SOTA；标准 harness 得分 63%，Provider Adapter harness 得分 99%，并在 96% 关卡上超过人类表现。","url":"https://www.aioga.com/news/cmtm2suk00171rooej14mutgj/","dateCreated":"2026-09-03T21:46:00.000Z","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":"@gdb) source article","url":"https://x.com/gdb/status/2095629409017614390","datePublished":"2026-09-03T21:46:00.000Z","provider":{"@type":"Organization","name":"@gdb)","url":"https://x.com/gdb/status/2095629409017614390"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.news/items/cmtm2suk00171rooej14mutgj","datePublished":"2026-09-03T21:46:00.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.news/items/cmtm2suk00171rooej14mutgj"}}],"aggregationSource":"@gdb)","originalPublisher":{"name":"@gdb)","url":"https://x.com/gdb/status/2095629409017614390"},"geoDeepAnswer":null,"article":{"id":"cmtm2suk00171rooej14mutgj","slug":"cmtm2suk00171rooej14mutgj","url":"https://www.aioga.com/news/cmtm2suk00171rooej14mutgj/","title":"Greg Brockman 转发：GPT-6 Astra 在 ARC-AGI-3 达到 SOTA，基准趋于饱和","title_en":"","summary":"Greg Brockman 转发 @arcprize 的评测称 OpenAI 的 GPT-6 Astra 在 ARC-AGI-3 上取得 SOTA，他称该基准已饱和。Astra 标准 harness 得分 63%，经新的 Provider Adapter harness 达 99%，在 96% 的 ARC-AGI-3 关卡上超越人类表现；排行榜图还显示更高推理层级通常成本更低，因为 Astra 用更少动作通关，减少模型调用和 token 数。","source":"@gdb)","sourceUrl":"https://x.com/gdb/status/2095629409017614390","aiHotUrl":"https://aihot.news/items/cmtm2suk00171rooej14mutgj","publishedAt":"2026-09-03T21:46:00.000Z","category":"行业动态","score":72,"selected":true,"articleBody":["Greg Brockman 转发 @arcprize 的评测称 OpenAI 的 GPT-6 Astra 在 ARC-AGI-3 上取得 SOTA，他称该基准已饱和。","Astra 标准 harness 得分 63%，经新的 Provider Adapter harness 达 99%，在 96% 的 ARC-AGI-3 关卡上超越人类表现；","排行榜图还显示更高推理层级通常成本更低，因为 Astra 用更少动作通关，减少模型调用和 token 数。"],"articleImages":[],"mediaStatus":"none","articleBodyZh":["Greg Brockman 转发：GPT-6 Astra 在 ARC-AGI-3 达到 SOTA，基准趋于饱和 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：Greg Brockman 转发 @arcprize 的评测称 OpenAI 的 GPT-6 Astra 在 ARC-AGI-3 上取得 SOTA，他称该基准已饱和。Astra 标准 harness 得分 63%，经新的 Provider Adapter harness 达 99%，在 96% 的 ARC-AGI-3 关卡上超越人类表现；排行榜图还显示更高推理层级通常成本更低，因为 Astra 用更少动作通关，减少模型调用和 token 数。","背景：背景分析：公司与行业类动态需要放在竞争格局、商业化路径、资本信号和监管环境中观察，单条公告不能代表最终结果。","Aioga 观察：Aioga 判断：这条动态更适合作为行业观察信号，当前信息足以建立线索，但不足以推导长期结论。","影响与后续：影响分析：对相关团队而言，短期应先核对来源、可用范围和实际成本，再判断是否值得接入或跟进。 后续观察：继续观察官方文件、合作落地、收入或用户信号、竞品动作和监管后续。"],"translationStatus":"","bodyOrigin":"social-summary","editorial":{"summary":"据 Greg Brockman 转发的 @arcprize 评测，材料称 OpenAI 的 GPT-6 Astra 在 ARC-AGI-3 取得 SOTA；标准 harness 得分 63%，Provider Adapter harness 得分 99%，并在 96% 关卡上超过人类表现。","background":"材料同时称，排行榜图显示更高推理层级通常成本更低，原因是 Astra 以更少动作通关，从而减少模型调用和 token 数；Brockman 还称该基准已饱和。","viewpoint":"Aioga 判断：这则信息的核心不只是榜单分数差异，也包括评测 harness 变化带来的结果差异。由于材料仅提供转述与摘要，SOTA、饱和及成本关系仍应结合完整评测规则核验。","implications":"可能影响：若相关评测结果可复现，ARC-AGI-3 的比较将需要同时关注 harness、动作数量、模型调用和 token 消耗；但单条材料不足以证明通用能力提升，也不代表所有任务成本都会下降。","nextStep":"后续观察：建议核对 ARC-AGI-3 排行榜、标准与 Provider Adapter harness 的定义、样本覆盖和复测结果，并持续观察“基准饱和”判断是否得到后续公开数据支持。","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":"2026-09-03T23:04:43.881Z","sourceHash":"69530744ba0200bf","review":{"approved":true,"groundedness":94,"clarity":92,"duplicationRisk":38,"blockingIssues":[],"notes":["内容准确区分了材料事实、观点和待核验事项。","“Aioga 判断”属于明确标注的观点，不构成观点冒充事实。","可将“在 96% 关卡上超过人类表现”进一步明确为材料所述评测结果，以避免被理解为一般性能力结论。"]},"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","@gdb)"],"translations":{"zh-CN":{"title":"Greg Brockman 转发：GPT-6 Astra 在 ARC-AGI-3 达到 SOTA，基准趋于饱和","summary":"Greg Brockman 转发 @arcprize 的评测称 OpenAI 的 GPT-6 Astra 在 ARC-AGI-3 上取得 SOTA，他称该基准已饱和。Astra 标准 harness 得分 63%，经新的 Provider Adapter harness 达 99%，在 96% 的 ARC-AGI-3 关卡上超越人类表现；排行榜图还显示更高推理层级通常成本更低，因为 Astra 用更少动作通关，减少模型调用和 token 数。","category":"行业动态","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Greg Brockman 转发：GPT-6 Astra 在 ARC-AGI-3 达到 SOTA，基准趋于饱和 - Aioga AI资讯","description":"Greg Brockman 转发 @arcprize 的评测称 OpenAI 的 GPT-6 Astra 在 ARC-AGI-3 上取得 SOTA，他称该基准已饱和。Astra 标准 harness 得分 63%，经新的 Provider Adapter harness 达 99%，在 96% 的 ARC-AGI-3 关卡上超越人类表现；排行榜图还显示更高推...","url":"https://www.aioga.com/news/cmtm2suk00171rooej14mutgj/","articleBody":["Greg Brockman 转发：GPT-6 Astra 在 ARC-AGI-3 达到 SOTA，基准趋于饱和 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：Greg Brockman 转发 @arcprize 的评测称 OpenAI 的 GPT-6 Astra 在 ARC-AGI-3 上取得 SOTA，他称该基准已饱和。Astra 标准 harness 得分 63%，经新的 Provider Adapter harness 达 99%，在 96% 的 ARC-AGI-3 关卡上超越人类表现；排行榜图还显示更高推理层级通常成本更低，因为 Astra 用更少动作通关，减少模型调用和 token 数。","背景：背景分析：公司与行业类动态需要放在竞争格局、商业化路径、资本信号和监管环境中观察，单条公告不能代表最终结果。","Aioga 观察：Aioga 判断：这条动态更适合作为行业观察信号，当前信息足以建立线索，但不足以推导长期结论。","影响与后续：影响分析：对相关团队而言，短期应先核对来源、可用范围和实际成本，再判断是否值得接入或跟进。 后续观察：继续观察官方文件、合作落地、收入或用户信号、竞品动作和监管后续。"]},"en":{"title":"Greg Brockman retweeted: GPT-6 Astra reached SOTA on ARC-AGI-3, benchmarking is approaching saturation","summary":"Greg Brockman retweeted @arcprize's evaluation stating that OpenAI's GPT-6 Astra achieved SOTA on ARC-AGI-3, and he mentioned that this benchmark is already saturated. Astra scored 63% on the standard harness, and with the new Provider Adapter harness, it reached 99%, surpassing human performance on 96% of ARC-AGI-3 levels. The leaderboard chart also shows that higher reasoning levels usually cost less, because Astra completes levels with fewer actions, reducing model calls and token consumption.","category":"Industry","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Greg Brockman retweeted: GPT-6 Astra reached SOTA on ARC-AGI-3, benchmarking is approaching saturation - Aioga AI News","description":"Greg Brockman retweeted @arcprize's evaluation stating that OpenAI's GPT-6 Astra achieved SOTA on ARC-AGI-3, and he mentioned that this benchmark is already saturated. Astra scored...","url":"https://www.aioga.com/en/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:22:56.409Z"},"ja":{"title":"グレッグ・ブロックマンが転送：GPT-6 Astra が ARC-AGI-3 で SOTA に達し、ベンチマークが飽和に近づく","summary":"Greg Brockman は @arcprize のレビューをリツイートし、OpenAI の GPT-6 Astra が ARC-AGI-3 で SOTA を達成したと述べ、彼はこのベンチマークは既に飽和していると述べています。Astra の標準ハーネスのスコアは 63% で、新しい Provider Adapter ハーネスでは 99% に達し、ARC-AGI-3 の 96% のステージで人間のパフォーマンスを上回りました。ランキング図は、推論レベルが高いほど通常コストが低いことも示しています。これは Astra がより少ないアクションでクリアし、モデル呼び出しとトークン数を減らしているためです。","category":"業界動向","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"グレッグ・ブロックマンが転送：GPT-6 Astra が ARC-AGI-3 で SOTA に達し、ベンチマークが飽和に近づく - Aioga AIニュース","description":"Greg Brockman は @arcprize のレビューをリツイートし、OpenAI の GPT-6 Astra が ARC-AGI-3 で SOTA を達成したと述べ、彼はこのベンチマークは既に飽和していると述べています。Astra の標準ハーネスのスコアは 63% で、新しい Provider Adapter ハーネスでは 99% に達し、ARC-...","url":"https://www.aioga.com/ja/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:23:03.582Z"},"ko":{"title":"Greg Brockman 전송: GPT-6 Astra가 ARC-AGI-3에서 SOTA에 도달, 벤치마크가 포화 상태에 가까워졌습니다","summary":"Greg Brockman이 @arcprize의 평가를 리트윗하며 OpenAI의 GPT-6 Astra가 ARC-AGI-3에서 SOTA를 달성했다고 밝혔다. 그는 이 벤치마크가 이미 포화 상태라고 언급했다. Astra 표준 하니스 점수는 63%였으나 새로운 Provider Adapter 하니스에서는 99%를 기록했으며, ARC-AGI-3 레벨의 96%에서 인간 성능을 능가했다. 순위표 그림은 더 높은 추론 단계일수록 일반적으로 비용이 낮다는 것을 보여주는데, Astra가 더 적은 액션으로 단계를 클리어해 모델 호출 수와 토큰 수를 줄였기 때문이다.","category":"업계 동향","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Greg Brockman 전송: GPT-6 Astra가 ARC-AGI-3에서 SOTA에 도달, 벤치마크가 포화 상태에 가까워졌습니다 - Aioga AI 뉴스","description":"Greg Brockman이 @arcprize의 평가를 리트윗하며 OpenAI의 GPT-6 Astra가 ARC-AGI-3에서 SOTA를 달성했다고 밝혔다. 그는 이 벤치마크가 이미 포화 상태라고 언급했다. Astra 표준 하니스 점수는 63%였으나 새로운 Provider Adapter 하니스에서는 99%를 기록했으며, AR...","url":"https://www.aioga.com/ko/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:23:38.112Z"},"es":{"title":"Greg Brockman retuiteó: GPT-6 Astra alcanza SOTA en ARC-AGI-3, los benchmarks tienden a saturarse","summary":"Greg Brockman compartió una evaluación de @arcprize que afirma que GPT-6 Astra de OpenAI ha alcanzado SOTA en ARC-AGI-3, y dijo que este benchmark ya está saturado. Astra obtuvo un puntaje del 63% en el harness estándar, y con el nuevo harness Provider Adapter alcanzó el 99%, superando el rendimiento humano en el 96% de los niveles de ARC-AGI-3; la gráfica del ranking también muestra que los niveles de razonamiento más altos suelen ser más económicos, ya que Astra completa los niveles con menos acciones, reduciendo las llamadas al modelo y la cantidad de tokens.","category":"Industria","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Greg Brockman retuiteó: GPT-6 Astra alcanza SOTA en ARC-AGI-3, los benchmarks tienden a saturarse - Aioga Noticias de IA","description":"Greg Brockman compartió una evaluación de @arcprize que afirma que GPT-6 Astra de OpenAI ha alcanzado SOTA en ARC-AGI-3, y dijo que este benchmark ya está saturado. Astra obtuvo un...","url":"https://www.aioga.com/es/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:23:33.289Z"},"fr":{"title":"Greg Brockman a relayé : GPT-6 Astra atteint SOTA sur ARC-AGI-3, le benchmark tend à saturer","summary":"Greg Brockman a relayé l'évaluation de @arcprize indiquant que GPT-6 Astra d'OpenAI a atteint le SOTA sur ARC-AGI-3, et il a déclaré que ce benchmark est saturé. Le score standard du harness d'Astra est de 63 %, et avec le nouveau Provider Adapter harness, il atteint 99 %, surpassant les performances humaines sur 96 % des niveaux d'ARC-AGI-3 ; le graphique du classement montre également que les niveaux de raisonnement plus élevés coûtent généralement moins cher, car Astra termine avec moins d'actions, réduisant le nombre d'appels au modèle et de tokens.","category":"Industrie","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Greg Brockman a relayé : GPT-6 Astra atteint SOTA sur ARC-AGI-3, le benchmark tend à saturer - Aioga Actualités IA","description":"Greg Brockman a relayé l'évaluation de @arcprize indiquant que GPT-6 Astra d'OpenAI a atteint le SOTA sur ARC-AGI-3, et il a déclaré que ce benchmark est saturé. Le score standard...","url":"https://www.aioga.com/fr/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:24:11.225Z"},"de":{"title":"Greg Brockman weitergeleitet: GPT-6 Astra erreicht SOTA bei ARC-AGI-3, Benchmark nähert sich der Sättigung","summary":"Greg Brockman hat die Bewertung von @arcprize weitergeleitet, in der angegeben wird, dass OpenAIs GPT-6 Astra in ARC-AGI-3 den SOTA erreicht hat. Er sagte, dass dieser Benchmark bereits gesättigt ist. Im Standard-Harness von Astra erzielte es 63 %, mit dem neuen Provider-Adapter-Harness 99 % und übertraf die menschliche Leistung in 96 % der ARC-AGI-3-Level; die Ranglisten-Grafik zeigt auch, dass höhere Denkebenen in der Regel kostengünstiger sind, da Astra mit weniger Aktionen durchkommt und so Modellaufrufe und Token reduziert.","category":"行业动态","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Greg Brockman weitergeleitet: GPT-6 Astra erreicht SOTA bei ARC-AGI-3, Benchmark nähert sich der Sättigung - Aioga KI-News","description":"Greg Brockman hat die Bewertung von @arcprize weitergeleitet, in der angegeben wird, dass OpenAIs GPT-6 Astra in ARC-AGI-3 den SOTA erreicht hat. Er sagte, dass dieser Benchmark be...","url":"https://www.aioga.com/de/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:24:11.621Z"},"pt-BR":{"title":"Greg Brockman compartilhou: GPT-6 Astra alcançou SOTA no ARC-AGI-3, os benchmarks estão se aproximando da saturação","summary":"Greg Brockman compartilhou a avaliação do @arcprize dizendo que o GPT-6 Astra da OpenAI atingiu SOTA no ARC-AGI-3, e ele afirmou que esse benchmark já está saturado. O score padrão do Astra harness foi de 63%, e com o novo Provider Adapter harness chegou a 99%, superando o desempenho humano em 96% dos níveis do ARC-AGI-3; o gráfico do ranking também mostra que níveis mais altos de raciocínio geralmente custam menos, porque o Astra completa com menos ações, reduzindo chamadas de modelo e número de tokens.","category":"行业动态","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Greg Brockman compartilhou: GPT-6 Astra alcançou SOTA no ARC-AGI-3, os benchmarks estão se aproximando da saturação - Aioga Notícias de IA","description":"Greg Brockman compartilhou a avaliação do @arcprize dizendo que o GPT-6 Astra da OpenAI atingiu SOTA no ARC-AGI-3, e ele afirmou que esse benchmark já está saturado. O score padrão...","url":"https://www.aioga.com/pt-BR/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:24:43.503Z"},"ru":{"title":"Грег Брокман ретвитнул: GPT-6 Astra достигает SOTA в ARC-AGI-3, бенчмарки приближаются к насыщению","summary":"Грег Брокман ретвитнул обзор @arcprize, в котором говорится, что GPT-6 Astra от OpenAI достигла SOTA на ARC-AGI-3, при этом он отметил, что этот бенчмарк уже насыщен. Стандартная система оценивания Astra набрала 63%, а с новым адаптером провайдера достигла 99%, превзойдя человеческие показатели на 96% уровней ARC-AGI-3; график рейтинга также показывает, что более высокий уровень рассуждений обычно стоит дешевле, так как Astra проходит уровни меньшим количеством действий, снижая количество вызовов модели и токенов.","category":"行业动态","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Грег Брокман ретвитнул: GPT-6 Astra достигает SOTA в ARC-AGI-3, бенчмарки приближаются к насыщению - Aioga Новости ИИ","description":"Грег Брокман ретвитнул обзор @arcprize, в котором говорится, что GPT-6 Astra от OpenAI достигла SOTA на ARC-AGI-3, при этом он отметил, что этот бенчмарк уже насыщен. Стандартная с...","url":"https://www.aioga.com/ru/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:24:45.217Z"},"ar":{"title":"غريغ بروكمان أعاد التغريد: GPT-6 Astra يصل إلى SOTA في ARC-AGI-3، والمعيار يقترب من التشبع","summary":"غريغ بروكمان أعاد تغريد تقييم @arcprize الذي يشير إلى أن GPT-6 Astra من OpenAI حقق أفضل أداء على ARC-AGI-3، وقال إن هذا المعيار قد وصل إلى التشبع. سجل معيار Astra العادي 63٪، وعبر معيار Provider Adapter الجديد وصل إلى 99٪، متفوقًا على الأداء البشري في 96٪ من مستويات ARC-AGI-3؛ كما تُظهر خريطة الترتيب أن المستويات الأعلى من الاستدلال عادة ما تكون أقل تكلفة، لأن Astra أنهى المهام بعدد أقل من الإجراءات، مما يقلل من استدعاءات النموذج وعدد الرموز (tokens).","category":"行业动态","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"غريغ بروكمان أعاد التغريد: GPT-6 Astra يصل إلى SOTA في ARC-AGI-3، والمعيار يقترب من التشبع - Aioga أخبار الذكاء الاصطناعي","description":"غريغ بروكمان أعاد تغريد تقييم @arcprize الذي يشير إلى أن GPT-6 Astra من OpenAI حقق أفضل أداء على ARC-AGI-3، وقال إن هذا المعيار قد وصل إلى التشبع. سجل معيار Astra العادي 63٪، وعبر...","url":"https://www.aioga.com/ar/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:25:21.411Z"},"hi":{"title":"ग्रेग ब्रॉकमैन ने रीट्वीट किया: GPT-6 Astra ने ARC-AGI-3 में SOTA हासिल किया, बेंचमार्क संतृप्ति की ओर बढ़ रहे हैं","summary":"Greg Brockman ने @arcprize की समीक्षा को साझा किया जिसमें कहा गया कि OpenAI का GPT-6 Astra ने ARC-AGI-3 पर SOTA हासिल किया है, उन्होंने कहा कि यह बेंचमार्क पहले ही संतृप्त हो चुका है। Astra मानक हार्नेस स्कोर 63% है, नए Provider Adapter हार्नेस के साथ यह 99% तक पहुंच गया, और ARC-AGI-3 के 96% स्तरों पर मानव प्रदर्शन को पार कर गया; लीडरबोर्ड ग्राफ़ यह भी दिखाता है कि उच्चतर तर्क स्तर आमतौर पर कम लागत वाला होता है, क्योंकि Astra कम क्रियाओं में स्तर पूरा करता है, मॉडल कॉल और टोकन की संख्या घटाता है।","category":"行业动态","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"ग्रेग ब्रॉकमैन ने रीट्वीट किया: GPT-6 Astra ने ARC-AGI-3 में SOTA हासिल किया, बेंचमार्क संतृप्ति की ओर बढ़ रहे हैं - Aioga AI समाचार","description":"Greg Brockman ने @arcprize की समीक्षा को साझा किया जिसमें कहा गया कि OpenAI का GPT-6 Astra ने ARC-AGI-3 पर SOTA हासिल किया है, उन्होंने कहा कि यह बेंचमार्क पहले ही संतृप्त हो चुका...","url":"https://www.aioga.com/hi/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:25:19.676Z"},"it":{"title":"Greg Brockman ha ritrasmesso: GPT-6 Astra ha raggiunto lo SOTA in ARC-AGI-3, il benchmark sta tendendo a saturarsi","summary":"Greg Brockman ha ritwittato la recensione di @arcprize, affermando che GPT-6 Astra di OpenAI ha raggiunto lo SOTA su ARC-AGI-3, e ha detto che questo benchmark è ormai saturo. Il punteggio standard dell'harness di Astra è del 63%, mentre con il nuovo Provider Adapter harness arriva al 99%, superando le performance umane nel 96% dei livelli di ARC-AGI-3; il grafico della classifica mostra anche che livelli di ragionamento più alti generalmente hanno costi più bassi, perché Astra completa i livelli con meno azioni, riducendo le chiamate al modello e il numero di token.","category":"行业动态","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Greg Brockman ha ritrasmesso: GPT-6 Astra ha raggiunto lo SOTA in ARC-AGI-3, il benchmark sta tendendo a saturarsi - Aioga Notizie IA","description":"Greg Brockman ha ritwittato la recensione di @arcprize, affermando che GPT-6 Astra di OpenAI ha raggiunto lo SOTA su ARC-AGI-3, e ha detto che questo benchmark è ormai saturo. Il p...","url":"https://www.aioga.com/it/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:25:58.739Z"},"nl":{"title":"Greg Brockman deelt: GPT-6 Astra bereikt SOTA op ARC-AGI-3, benchmarks naderen verzadiging","summary":"Greg Brockman deelt een beoordeling van @arcprize waarin staat dat OpenAI's GPT-6 Astra SOTA heeft behaald op ARC-AGI-3, en hij zegt dat deze benchmark al verzadigd is. Astra behaalt 63% op de standaard harness, en met de nieuwe Provider Adapter harness bereikt het 99%, waarmee het 96% van de ARC-AGI-3 niveaus beter presteert dan mensen; de ranglijstgrafiek laat ook zien dat een hoger redeneerniveau doorgaans lagere kosten met zich meebrengt, omdat Astra met minder acties door het niveau komt, waardoor het modelgebruik en het aantal tokens wordt verminderd.","category":"行业动态","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Greg Brockman deelt: GPT-6 Astra bereikt SOTA op ARC-AGI-3, benchmarks naderen verzadiging - Aioga AI-nieuws","description":"Greg Brockman deelt een beoordeling van @arcprize waarin staat dat OpenAI's GPT-6 Astra SOTA heeft behaald op ARC-AGI-3, en hij zegt dat deze benchmark al verzadigd is. Astra behaa...","url":"https://www.aioga.com/nl/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:26:01.450Z"},"tr":{"title":"Greg Brockman aktardı: GPT-6 Astra ARC-AGI-3'te SOTA'ya ulaştı, kıyaslama doygunluğa doğru ilerliyor","summary":"Greg Brockman, @arcprize'ın değerlendirmesini aktararak, OpenAI'nin GPT-6 Astra'sının ARC-AGI-3 üzerinde SOTA elde ettiğini belirtti ve bu kıstasın artık doygunluğa ulaştığını söyledi. Astra standart harness puanı %63, yeni Provider Adapter harness ile %99'a ulaşıyor ve ARC-AGI-3 seviyelerinin %96'sında insan performansını aşıyor; sıralama grafiği ayrıca daha yüksek akıl yürütme seviyelerinin genellikle maliyetinin daha düşük olduğunu gösteriyor, çünkü Astra daha az hareketle geçiş yaparak model çağrılarını ve token sayısını azaltıyor.","category":"行业动态","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Greg Brockman aktardı: GPT-6 Astra ARC-AGI-3'te SOTA'ya ulaştı, kıyaslama doygunluğa doğru ilerliyor - Aioga AI Haberleri","description":"Greg Brockman, @arcprize'ın değerlendirmesini aktararak, OpenAI'nin GPT-6 Astra'sının ARC-AGI-3 üzerinde SOTA elde ettiğini belirtti ve bu kıstasın artık doygunluğa ulaştığını söyl...","url":"https://www.aioga.com/tr/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:26:33.870Z"},"vi":{"title":"Greg Brockman chuyển tiếp: GPT-6 Astra đạt SOTA tại ARC-AGI-3, chuẩn mực đang dần bão hòa","summary":"Greg Brockman chuyển tiếp đánh giá của @arcprize cho biết GPT-6 Astra của OpenAI đã đạt SOTA trên ARC-AGI-3, ông cho rằng chuẩn đánh giá này đã bão hòa. Điểm chuẩn chuẩn Astra đạt 63%, nhưng thông qua harness Provider Adapter mới đạt 99%, vượt hiệu suất con người ở 96% các cấp độ của ARC-AGI-3; biểu đồ bảng xếp hạng cũng cho thấy mức suy luận cao hơn thường tốn ít chi phí hơn, vì Astra hoàn thành cấp độ với ít hành động hơn, giảm số lần gọi mô hình và token.","category":"行业动态","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Greg Brockman chuyển tiếp: GPT-6 Astra đạt SOTA tại ARC-AGI-3, chuẩn mực đang dần bão hòa - Tin tức AI Aioga","description":"Greg Brockman chuyển tiếp đánh giá của @arcprize cho biết GPT-6 Astra của OpenAI đã đạt SOTA trên ARC-AGI-3, ông cho rằng chuẩn đánh giá này đã bão hòa. Điểm chuẩn chuẩn Astra đạt...","url":"https://www.aioga.com/vi/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:26:35.563Z"},"id":{"title":"Greg Brockman meneruskan: GPT-6 Astra mencapai SOTA di ARC-AGI-3, tolok ukur mendekati kejenuhan","summary":"Greg Brockman meneruskan ulasan @arcprize yang menyatakan OpenAI GPT-6 Astra mencapai SOTA pada ARC-AGI-3, dia mengatakan tolok ukur ini telah jenuh. Skor standard harness Astra adalah 63%, melalui harness Provider Adapter baru mencapai 99%, melampaui kinerja manusia di 96% level ARC-AGI-3; grafik peringkat juga menunjukkan tingkat penalaran yang lebih tinggi biasanya biayanya lebih rendah, karena Astra menyelesaikan level dengan tindakan lebih sedikit, mengurangi panggilan model dan jumlah token.","category":"行业动态","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Greg Brockman meneruskan: GPT-6 Astra mencapai SOTA di ARC-AGI-3, tolok ukur mendekati kejenuhan - Berita AI Aioga","description":"Greg Brockman meneruskan ulasan @arcprize yang menyatakan OpenAI GPT-6 Astra mencapai SOTA pada ARC-AGI-3, dia mengatakan tolok ukur ini telah jenuh. Skor standard harness Astra ad...","url":"https://www.aioga.com/id/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:27:08.085Z"},"th":{"title":"Greg Brockman รีทวีต: GPT-6 Astra บรรลุ SOTA ใน ARC-AGI-3 เกณฑ์เริ่มเข้าสู่ความอิ่มตัว","summary":"Greg Brockman แชร์บทวิจารณ์ของ @arcprize ว่า GPT-6 Astra ของ OpenAI ทำผลงานเป็น SOTA ใน ARC-AGI-3 เขากล่าวว่ามาตรฐานนี้ถึงจุดอิ่มตัวแล้ว คะแนนมาตรฐานของ Astra อยู่ที่ 63% แต่เมื่อใช้ Provider Adapter harness ใหม่ ทำได้ถึง 99% และสามารถทำได้ดีกว่ามนุษย์ใน 96% ของด่าน ARC-AGI-3; กราฟอันดับยังแสดงว่าระดับการวิเคราะห์ที่สูงขึ้นมักมีต้นทุนต่ำกว่า เพราะ Astra ใช้การกระทำน้อยลงในการผ่านด่าน ลดการเรียกใช้งานโมเดลและจำนวนโทเค็น","category":"行业动态","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Greg Brockman รีทวีต: GPT-6 Astra บรรลุ SOTA ใน ARC-AGI-3 เกณฑ์เริ่มเข้าสู่ความอิ่มตัว - ข่าว AI Aioga","description":"Greg Brockman แชร์บทวิจารณ์ของ @arcprize ว่า GPT-6 Astra ของ OpenAI ทำผลงานเป็น SOTA ใน ARC-AGI-3 เขากล่าวว่ามาตรฐานนี้ถึงจุดอิ่มตัวแล้ว คะแนนมาตรฐานของ Astra อยู่ที่ 63% แต่เมื่อใ...","url":"https://www.aioga.com/th/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:27:13.040Z"},"pl":{"title":"Greg Brockman przekazuje dalej: GPT-6 Astra osiągnął SOTA w ARC-AGI-3, benchmark zbliża się do nasycenia","summary":"Greg Brockman udostępnił w sieci recenzję @arcprize, w której stwierdzono, że GPT-6 Astra od OpenAI osiągnął SOTA w ARC-AGI-3, i powiedział, że ten benchmark jest już nasycony. Standardowy harness Asty uzyskał wynik 63%, a dzięki nowemu harness Provider Adapter osiągnął 99%, przekraczając wyniki ludzkie w 96% poziomów ARC-AGI-3; wykres rankingowy pokazuje również, że wyższe poziomy wnioskowania zazwyczaj kosztują mniej, ponieważ Astra przechodzi poziomy za pomocą mniejszej liczby działań, zmniejszając liczbę wywołań modelu i tokenów.","category":"行业动态","source":"@gdb)","aggregationSource":"@gdb)","pageTitle":"Greg Brockman przekazuje dalej: GPT-6 Astra osiągnął SOTA w ARC-AGI-3, benchmark zbliża się do nasycenia - Aioga Wiadomości AI","description":"Greg Brockman udostępnił w sieci recenzję @arcprize, w której stwierdzono, że GPT-6 Astra od OpenAI osiągnął SOTA w ARC-AGI-3, i powiedział, że ten benchmark jest już nasycony. Sta...","url":"https://www.aioga.com/pl/news/cmtm2suk00171rooej14mutgj/","contentTranslated":true,"sourceHash":"a33d2e2379d264e1","translatedAt":"2026-09-03T22:27:49.354Z"}},"evidenceTier":"social-signal","reviewStatus":"editorial-selected","indexable":true,"editorialCover":"/page-visuals/topic-timeline.png"}}