{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-07-28T05:20:57.982Z","headline":"AREX：面向深度研究的递归自改进智能体","description":"AREX 是一系列递归自改进（RSI）深度研究智能体，通过内层研究循环收集证据、外层自改进循环逐约束审计答案并启动针对性研究。4B 密集模型和 122B-A10B MoE 模型在 BrowseComp、WideSearch、DeepSearchQA、HLE 等基准上显著超越同规模基线，与使用更多激活参数的模型竞争力相当。","url":"https://www.aioga.com/news/cmryek7m200akrolgoaqospqi/","mainEntityOfPage":"https://www.aioga.com/news/cmryek7m200akrolgoaqospqi/","datePublished":"2026-07-23T00:00:00.000Z","dateModified":"2026-07-23T00:00:00.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://arxiv.org/abs/2607.21461","https://aihot.virxact.com/items/cmryek7m200akrolgoaqospqi"],"canonicalUrl":"https://www.aioga.com/news/cmryek7m200akrolgoaqospqi/","directAnswer":{"@type":"Answer","text":"AREX 是一系列面向深度研究的递归自改进智能体，结合内层证据收集与外层逐约束审计，并可根据审计结果启动针对性研究。","url":"https://www.aioga.com/news/cmryek7m200akrolgoaqospqi/","dateCreated":"2026-07-23T00:00:00.000Z","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":"arXiv source article","url":"https://arxiv.org/abs/2607.21461","datePublished":"2026-07-23T00:00:00.000Z","provider":{"@type":"Organization","name":"arXiv","url":"https://arxiv.org/abs/2607.21461"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.virxact.com/items/cmryek7m200akrolgoaqospqi","datePublished":"2026-07-23T00:00:00.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.virxact.com/items/cmryek7m200akrolgoaqospqi"}}],"aggregationSource":"HuggingFace Daily Papers（社区热门论文）","originalPublisher":{"name":"arXiv","url":"https://arxiv.org/abs/2607.21461"},"article":{"id":"cmryek7m200akrolgoaqospqi","slug":"cmryek7m200akrolgoaqospqi","url":"https://www.aioga.com/news/cmryek7m200akrolgoaqospqi/","title":"AREX：面向深度研究的递归自改进智能体","title_en":"AREX： Towards a Recursively Self-Improving Agent for Deep Research","summary":"AREX 是一系列递归自改进（RSI）深度研究智能体，通过内层研究循环收集证据、外层自改进循环逐约束审计答案并启动针对性研究。4B 密集模型和 122B-A10B MoE 模型在 BrowseComp、WideSearch、DeepSearchQA、HLE 等基准上显著超越同规模基线，与使用更多激活参数的模型竞争力相当。","source":"HuggingFace Daily Papers（社区热门论文）","sourceUrl":"https://arxiv.org/abs/2607.21461","aiHotUrl":"https://aihot.virxact.com/items/cmryek7m200akrolgoaqospqi","publishedAt":"2026-07-23T00:00:00.000Z","category":"论文研究","score":78,"selected":true,"articleBody":["AREX 是一系列递归自改进（RSI）深度研究智能体，通过内层研究循环收集证据、外层自改进循环逐约束审计答案并启动针对性研究。","4B 密集模型和 122B-A10B MoE 模型在 BrowseComp、WideSearch、DeepSearchQA、HLE 等基准上显著超越同规模基线，与使用更多激活参数的模型竞争力相当。"],"articleImages":[],"mediaStatus":"none","articleBodyZh":[],"translationStatus":"","bodyOrigin":"summary-fallback","editorial":{"summary":"AREX 是一系列面向深度研究的递归自改进智能体，结合内层证据收集与外层逐约束审计，并可根据审计结果启动针对性研究。","background":"公开材料将其归入论文研究，来源为 HuggingFace Daily Papers 社区热门论文。研究涉及 4B 密集模型与 122B-A10B MoE 模型。","viewpoint":"Aioga 判断，AREX 的核心关注点不是单次生成答案，而是把证据收集、约束审计和补充研究组织为递归循环，以改善深度研究过程。","implications":"材料称，两种模型在 BrowseComp、WideSearch、DeepSearchQA、HLE 等基准上显著超越同规模基线，并可与使用更多激活参数的模型竞争。","nextStep":"值得关注后续公开材料是否披露更完整的实验设置、逐项基准结果、审计约束设计与针对性研究机制，以便进一步核验效果和适用边界。","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":"2026-07-26T02:50:12.394Z","sourceHash":"4f8f967d4083a530","review":{"approved":true,"groundedness":96,"clarity":94,"duplicationRisk":18,"blockingIssues":[],"notes":["viewpoint 明确以“Aioga 判断”标示分析性观点，未冒充来源事实。","nextStep 属于后续关注建议，并未断言来源材料已经披露相关细节。","“两种模型”可选地改为“4B 密集模型和 122B-A10B MoE 模型”，以减少指代歧义，但不影响审核通过。"]},"validation":{"passed":true,"mode":"ai-auto","revisions":0,"checks":["schema","length","source-attribution","low-source-overlap","no-html","independent-ai-review"]}},"tags":["论文研究","HuggingFace Daily Papers（社区热门论文）"],"translations":{"zh-CN":{"title":"AREX：面向深度研究的递归自改进智能体","summary":"AREX 是一系列递归自改进（RSI）深度研究智能体，通过内层研究循环收集证据、外层自改进循环逐约束审计答案并启动针对性研究。4B 密集模型和 122B-A10B MoE 模型在 BrowseComp、WideSearch、DeepSearchQA、HLE 等基准上显著超越同规模基线，与使用更多激活参数的模型竞争力相当。","category":"论文研究","source":"arXiv","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX：面向深度研究的递归自改进智能体 - Aioga AI资讯","description":"AREX 是一系列递归自改进（RSI）深度研究智能体，通过内层研究循环收集证据、外层自改进循环逐约束审计答案并启动针对性研究。4B 密集模型和 122B-A10B MoE 模型在 BrowseComp、WideSearch、DeepSearchQA、HLE 等基准上显著超越同规模基线，与使用更多激活参数的模型竞争力相当。","url":"https://www.aioga.com/news/cmryek7m200akrolgoaqospqi/"},"en":{"title":"AREX: A Recursively Self-Improving Agent for Deep Research","summary":"AREX is a series of recursive self-improving (RSI) deep research agents that collect evidence through inner research cycles and audit answers with constraints through outer self-improvement cycles, initiating targeted research. The 4B dense model and 122B-A10B MoE model significantly outperform same-scale baselines on benchmarks such as BrowseComp, WideSearch, DeepSearchQA, and HLE, and are competitive with models using more activated parameters.","category":"Research","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: A Recursively Self-Improving Agent for Deep Research - Aioga AI News","description":"AREX is a series of recursive self-improving (RSI) deep research agents that collect evidence through inner research cycles and audit answers with constraints through outer self-im...","url":"https://www.aioga.com/en/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:36:42.368Z"},"ja":{"title":"AREX：深層研究向けの再帰的自己改良エージェント","summary":"AREX は一連の再帰的自己改善（RSI）深層研究エージェントであり、内層の研究サイクルで証拠を収集し、外層の自己改善サイクルで制約に基づき回答を監査し、ターゲットを絞った研究を開始します。4B 集約モデルと 122B-A10B MoE モデルは、BrowseComp、WideSearch、DeepSearchQA、HLE などのベンチマークで同規模のベースラインを大きく上回り、より多くのアクティベーションパラメータを使用するモデルと同等の競争力を示します。","category":"論文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX：深層研究向けの再帰的自己改良エージェント - Aioga AIニュース","description":"AREX は一連の再帰的自己改善（RSI）深層研究エージェントであり、内層の研究サイクルで証拠を収集し、外層の自己改善サイクルで制約に基づき回答を監査し、ターゲットを絞った研究を開始します。4B 集約モデルと 122B-A10B MoE モデルは、BrowseComp、WideSearch、DeepSearchQA、HLE などのベンチマークで同規模のベース...","url":"https://www.aioga.com/ja/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:36:47.240Z"},"ko":{"title":"AREX: 깊이 연구를 위한 재귀 자기개선 에이전트","summary":"AREX는 일련의 재귀적 자기 개선(RSI) 심층 연구 에이전트로, 내부 연구 루프를 통해 증거를 수집하고 외부 자기 개선 루프에서 제약 조건에 따라 답변을 감사하며 목표 지향적 연구를 시작합니다. 4B 집약 모델과 122B-A10B MoE 모델은 BrowseComp, WideSearch, DeepSearchQA, HLE 등 벤치마크에서 동급 크기 기반 모델을 크게 능가하며, 더 많은 활성화 파라미터를 사용하는 모델과 경쟁할 만한 성능을 보여줍니다.","category":"연구","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: 깊이 연구를 위한 재귀 자기개선 에이전트 - Aioga AI 뉴스","description":"AREX는 일련의 재귀적 자기 개선(RSI) 심층 연구 에이전트로, 내부 연구 루프를 통해 증거를 수집하고 외부 자기 개선 루프에서 제약 조건에 따라 답변을 감사하며 목표 지향적 연구를 시작합니다. 4B 집약 모델과 122B-A10B MoE 모델은 BrowseComp, WideSearch, DeepSearchQA, HLE...","url":"https://www.aioga.com/ko/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:37:28.071Z"},"es":{"title":"AREX: Agente Inteligente de Auto-mejora Recursiva para Investigación Profunda","summary":"AREX es una serie de agentes inteligentes de investigación profunda de mejora recursiva (RSI), que recopilan evidencia a través de ciclos de investigación internos y auditan respuestas de manera gradual en ciclos externos de auto-mejora, iniciando investigaciones específicas según las necesidades. Los modelos intensivos de 4B y el modelo MoE 122B-A10B superan significativamente las líneas base de tamaño similar en los estándares BrowseComp, WideSearch, DeepSearchQA, HLE, y son comparables en competitividad con modelos que utilizan más parámetros de activación.","category":"Investigación","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: Agente Inteligente de Auto-mejora Recursiva para Investigación Profunda - Aioga Noticias de IA","description":"AREX es una serie de agentes inteligentes de investigación profunda de mejora recursiva (RSI), que recopilan evidencia a través de ciclos de investigación internos y auditan respue...","url":"https://www.aioga.com/es/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:37:26.061Z"},"fr":{"title":"AREX : Agent intelligent récursivement auto-amélioré pour la recherche approfondie","summary":"AREX est une série d'agents intelligents de recherche approfondie récursive auto-améliorante (RSI), qui collectent des preuves à travers des cycles de recherche internes et auditent les réponses selon des contraintes via des cycles d'auto-amélioration externes tout en lançant des recherches ciblées. Les modèles intensifs 4B et les modèles 122B-A10B MoE surpassent de manière significative les bases de référence de même taille sur BrowseComp, WideSearch, DeepSearchQA, HLE, et sont comparables aux modèles utilisant plus de paramètres d'activation.","category":"Recherche","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX : Agent intelligent récursivement auto-amélioré pour la recherche approfondie - Aioga Actualités IA","description":"AREX est une série d'agents intelligents de recherche approfondie récursive auto-améliorante (RSI), qui collectent des preuves à travers des cycles de recherche internes et auditen...","url":"https://www.aioga.com/fr/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:38:04.414Z"},"de":{"title":"AREX: Rekursive selbstverbessernde Agenten für tiefgehende Forschung","summary":"AREX ist eine Reihe von rekursiv selbstverbessernden (RSI) tiefen Forschungsagenten, die durch innere Forschungsschleifen Beweise sammeln, in äußeren Selbstverbesserungsschleifen die Antworten anhand von Einschränkungen prüfen und gezielte Forschung initiieren. Die 4B-Dichte-Modelle und das 122B-A10B-MoE-Modell übertreffen in Benchmarks wie BrowseComp, WideSearch, DeepSearchQA und HLE deutlich gleich große Basismodelle und sind mit Modellen, die mehr Aktivierungsparameter verwenden, konkurrenzfähig.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: Rekursive selbstverbessernde Agenten für tiefgehende Forschung - Aioga KI-News","description":"AREX ist eine Reihe von rekursiv selbstverbessernden (RSI) tiefen Forschungsagenten, die durch innere Forschungsschleifen Beweise sammeln, in äußeren Selbstverbesserungsschleifen d...","url":"https://www.aioga.com/de/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:38:06.576Z"},"pt-BR":{"title":"AREX: Agente Recursivo de Autoaperfeiçoamento para Pesquisa Profunda","summary":"AREX é uma série de agentes inteligentes de pesquisa profunda recursiva de autoaperfeiçoamento (RSI), que coletam evidências por meio de ciclos de pesquisa internos e auditam respostas de acordo com restrições através de ciclos externos de autoaperfeiçoamento, iniciando pesquisas direcionadas. Os modelos intensivos 4B e 122B-A10B MoE superam significativamente as linhas de base de mesmo tamanho nos benchmarks BrowseComp, WideSearch, DeepSearchQA, HLE, etc., sendo competitivos com modelos que utilizam mais parâmetros ativos.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: Agente Recursivo de Autoaperfeiçoamento para Pesquisa Profunda - Aioga Notícias de IA","description":"AREX é uma série de agentes inteligentes de pesquisa profunda recursiva de autoaperfeiçoamento (RSI), que coletam evidências por meio de ciclos de pesquisa internos e auditam respo...","url":"https://www.aioga.com/pt-BR/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:38:46.565Z"},"ru":{"title":"AREX: Рекурсивно самоулучшающийся агент для углубленных исследований","summary":"AREX — это серия рекурсивно самоулучшающихся (RSI) агентов глубокого исследования, которые собирают доказательства с помощью внутреннего исследовательского цикла, а внешний цикл самоулучшения ограниченно проверяет ответы и инициирует целенаправленные исследования. 4B плотная модель и 122B-A10B MoE модель значительно превосходят базовые модели того же размера на бенчмарках BrowseComp, WideSearch, DeepSearchQA, HLE и других, сравнимы с моделями, которые используют больше активируемых параметров.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: Рекурсивно самоулучшающийся агент для углубленных исследований - Aioga Новости ИИ","description":"AREX — это серия рекурсивно самоулучшающихся (RSI) агентов глубокого исследования, которые собирают доказательства с помощью внутреннего исследовательского цикла, а внешний цикл са...","url":"https://www.aioga.com/ru/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:38:46.655Z"},"ar":{"title":"AREX: وكيل ذكي ذو تحسين ذاتي متكرر موجه نحو البحث العميق","summary":"AREX هو سلسلة من الوكلاء الأذكياء المتعمقين الذين يستخدمون التحسين الذاتي التكراري (RSI)، حيث يجمعون الأدلة من خلال دورة البحث الداخلية، ويقومون بمراجعة الإجابات وفق القيود عبر دورة التحسين الذاتي الخارجية، ويطلقون بحوثًا مستهدفة. تتجاوز نماذج 4B المكثفة ونماذج 122B-A10B MoE بشكل ملحوظ الخطوط الأساسية المماثلة في معايير مثل BrowseComp وWideSearch وDeepSearchQA وHLE، وتنافس النماذج التي تستخدم المزيد من معلمات التفعيل.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: وكيل ذكي ذو تحسين ذاتي متكرر موجه نحو البحث العميق - Aioga أخبار الذكاء الاصطناعي","description":"AREX هو سلسلة من الوكلاء الأذكياء المتعمقين الذين يستخدمون التحسين الذاتي التكراري (RSI)، حيث يجمعون الأدلة من خلال دورة البحث الداخلية، ويقومون بمراجعة الإجابات وفق القيود عبر دور...","url":"https://www.aioga.com/ar/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:39:28.551Z"},"hi":{"title":"AREX: गहन अनुसंधान के लिए पुनरावर्ती स्व-सुधार एजेंट","summary":"AREX एक श्रृंखला है जो पुनरावर्ती स्व-सुधार (RSI) गहन अनुसंधान एजेंट हैं, जो आंतरिक अनुसंधान चक्र के माध्यम से साक्ष्य एकत्र करते हैं, बाहरी स्व-सुधार चक्र के माध्यम से उत्तरों की प्रतिबंधित ऑडिटिंग करते हैं और लक्षित अनुसंधान को शुरू करते हैं। 4B घनी मॉडल और 122B-A10B MoE मॉडल BrowseComp, WideSearch, DeepSearchQA, HLE जैसे बेंचमार्क पर समान पैमाने के बेसलाइन को महत्वपूर्ण रूप से पार कर जाते हैं और अधिक सक्रिय पैरामीटर वाले मॉडलों के बराबर प्रतिस्पर्धात्मक होते हैं।","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: गहन अनुसंधान के लिए पुनरावर्ती स्व-सुधार एजेंट - Aioga AI समाचार","description":"AREX एक श्रृंखला है जो पुनरावर्ती स्व-सुधार (RSI) गहन अनुसंधान एजेंट हैं, जो आंतरिक अनुसंधान चक्र के माध्यम से साक्ष्य एकत्र करते हैं, बाहरी स्व-सुधार चक्र के माध्यम से उत्तरों की...","url":"https://www.aioga.com/hi/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:39:33.937Z"},"it":{"title":"AREX: Agente ricorsivo auto-migliorante per la ricerca approfondita","summary":"AREX è una serie di agenti di ricerca profonda a miglioramento ricorsivo (RSI) che raccolgono prove attraverso cicli di ricerca interni e auditano le risposte tramite cicli di auto-miglioramento esterni per avviare ricerche mirate. I modelli intensivi da 4B e i modelli MoE 122B-A10B superano significativamente le baseline della stessa scala su benchmark come BrowseComp, WideSearch, DeepSearchQA, HLE, risultando competitivi con modelli che utilizzano un maggior numero di parametri attivi.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: Agente ricorsivo auto-migliorante per la ricerca approfondita - Aioga Notizie IA","description":"AREX è una serie di agenti di ricerca profonda a miglioramento ricorsivo (RSI) che raccolgono prove attraverso cicli di ricerca interni e auditano le risposte tramite cicli di auto...","url":"https://www.aioga.com/it/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:40:12.862Z"},"nl":{"title":"AREX: een recursief zelfverbeterend agent gericht op diepgaand onderzoek","summary":"AREX is een reeks recursief zelfverbeterende (RSI) diepgaande onderzoeksagenten, die bewijs verzamelen via een innerlijke onderzoeksloop en antwoorden controleren en gerichte studies initiëren via een buitenste zelfverbeteringsloop. Het 4B-dichtheidsmodel en het 122B-A10B MoE-model overtreffen significant de basismodellen van dezelfde omvang op benchmarks zoals BrowseComp, WideSearch, DeepSearchQA, HLE, en zijn concurrerend met modellen die meer actieve parameters gebruiken.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: een recursief zelfverbeterend agent gericht op diepgaand onderzoek - Aioga AI-nieuws","description":"AREX is een reeks recursief zelfverbeterende (RSI) diepgaande onderzoeksagenten, die bewijs verzamelen via een innerlijke onderzoeksloop en antwoorden controleren en gerichte studi...","url":"https://www.aioga.com/nl/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:40:12.580Z"},"tr":{"title":"AREX: Derinlemesine Araştırmaya Yönelik Özyinelemeli Kendini Geliştiren Ajan","summary":"AREX, iç döngü araştırmalarıyla kanıt toplayan ve dış döngü kendini geliştirme ile yanıtları kısıtlama denetiminden geçiren ve hedefli araştırmaları başlatan bir dizi özyinelemeli kendini geliştiren (RSI) derin araştırma ajanıdır. 4B yoğun model ve 122B-A10B MoE model, BrowseComp, WideSearch, DeepSearchQA, HLE gibi kıyaslamalarda aynı ölçekli temelleri önemli ölçüde aşmış ve daha fazla aktif parametre kullanan modellerle rekabet edebilir düzeyde performans göstermiştir.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: Derinlemesine Araştırmaya Yönelik Özyinelemeli Kendini Geliştiren Ajan - Aioga AI Haberleri","description":"AREX, iç döngü araştırmalarıyla kanıt toplayan ve dış döngü kendini geliştirme ile yanıtları kısıtlama denetiminden geçiren ve hedefli araştırmaları başlatan bir dizi özyinelemeli...","url":"https://www.aioga.com/tr/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:40:56.295Z"},"vi":{"title":"AREX: Tác nhân tự cải tiến hồi quy hướng đến nghiên cứu sâu","summary":"AREX là một loạt các tác nhân nghiên cứu sâu cải tiến theo hồi quy (RSI), thông qua vòng nghiên cứu nội bộ để thu thập bằng chứng và vòng tự cải tiến bên ngoài để kiểm tra câu trả lời theo các ràng buộc và khởi động nghiên cứu có mục tiêu. Các mô hình mật độ 4B và mô hình 122B-A10B MoE vượt trội đáng kể so với các chuẩn mực cùng kích thước như BrowseComp, WideSearch, DeepSearchQA, HLE, và cạnh tranh tương đương với các mô hình sử dụng nhiều tham số kích hoạt hơn.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: Tác nhân tự cải tiến hồi quy hướng đến nghiên cứu sâu - Tin tức AI Aioga","description":"AREX là một loạt các tác nhân nghiên cứu sâu cải tiến theo hồi quy (RSI), thông qua vòng nghiên cứu nội bộ để thu thập bằng chứng và vòng tự cải tiến bên ngoài để kiểm tra câu trả...","url":"https://www.aioga.com/vi/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:40:51.575Z"},"id":{"title":"AREX: Agen Cerdas Rekursif yang Memperbaiki Diri Sendiri untuk Penelitian Mendalam","summary":"AREX adalah serangkaian agen penelitian cerdas dengan perbaikan diri rekursif (RSI), yang mengumpulkan bukti melalui siklus penelitian internal, meninjau jawaban secara bertahap melalui siklus perbaikan diri eksternal, dan memulai penelitian yang ditargetkan. Model padat 4B dan model MoE 122B-A10B secara signifikan melampaui tolok ukur sekelasnya seperti BrowseComp, WideSearch, DeepSearchQA, HLE, dan setara kompetitif dengan model yang menggunakan lebih banyak parameter aktivasi.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: Agen Cerdas Rekursif yang Memperbaiki Diri Sendiri untuk Penelitian Mendalam - Berita AI Aioga","description":"AREX adalah serangkaian agen penelitian cerdas dengan perbaikan diri rekursif (RSI), yang mengumpulkan bukti melalui siklus penelitian internal, meninjau jawaban secara bertahap me...","url":"https://www.aioga.com/id/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:41:33.572Z"},"th":{"title":"AREX: ตัวแทนอัจฉริยะที่ปรับปรุงตัวเองแบบวนซ้ำเพื่อการวิจัยเชิงลึก","summary":"AREX คือกลุ่มเอเจนต์อัจฉริยะที่ทำการศึกษาเชิงลึกแบบปรับปรุงตัวเองอย่างต่อเนื่อง (RSI) โดยเก็บรวบรวมหลักฐานผ่านวงจรการวิจัยภายใน และตรวจสอบคำตอบด้วยวงจรการปรับปรุงตัวเองภายนอกพร้อมเริ่มการวิจัยเฉพาะ จุด โมเดลเข้มข้น 4B และโมเดล 122B-A10B MoE ทำได้ดีกว่าในเกณฑ์มาตรฐาน BrowseComp, WideSearch, DeepSearchQA, HLE อย่างชัดเจนเมื่อเทียบกับฐานมาตรฐานขนาดเดียวกัน และสามารถแข่งขันกับโมเดลที่ใช้พารามิเตอร์มากกว่าได้","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: ตัวแทนอัจฉริยะที่ปรับปรุงตัวเองแบบวนซ้ำเพื่อการวิจัยเชิงลึก - ข่าว AI Aioga","description":"AREX คือกลุ่มเอเจนต์อัจฉริยะที่ทำการศึกษาเชิงลึกแบบปรับปรุงตัวเองอย่างต่อเนื่อง (RSI) โดยเก็บรวบรวมหลักฐานผ่านวงจรการวิจัยภายใน และตรวจสอบคำตอบด้วยวงจรการปรับปรุงตัวเองภายนอกพร้อมเ...","url":"https://www.aioga.com/th/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:41:39.727Z"},"pl":{"title":"AREX: Agent rekurencyjnie samodoskonalący się do badań głębokich","summary":"AREX to seria głęboko badawczych agentów inteligentnych z rekurencyjną samodoskonalącą się (RSI) strukturą, którzy gromadzą dowody w wewnętrznych cyklach badawczych, a w zewnętrznych cyklach samodoskonalenia ograniczają odpowiedzi zgodnie z zasadami audytu i inicjują ukierunkowane badania. Modele 4B intensywne oraz 122B-A10B MoE wyraźnie przewyższają linię bazową o tym samym rozmiarze w benchmarkach takich jak BrowseComp, WideSearch, DeepSearchQA, HLE i innych, wykazując konkurencyjność w porównaniu z modelami wykorzystującymi więcej parametrów aktywacji.","category":"论文研究","source":"HuggingFace Daily Papers（社区热门论文）","aggregationSource":"HuggingFace Daily Papers（社区热门论文）","pageTitle":"AREX: Agent rekurencyjnie samodoskonalący się do badań głębokich - Aioga Wiadomości AI","description":"AREX to seria głęboko badawczych agentów inteligentnych z rekurencyjną samodoskonalącą się (RSI) strukturą, którzy gromadzą dowody w wewnętrznych cyklach badawczych, a w zewnętrzny...","url":"https://www.aioga.com/pl/news/cmryek7m200akrolgoaqospqi/","contentTranslated":true,"sourceHash":"47e817569671e363","translatedAt":"2026-07-26T01:42:24.727Z"}}}}