{"@context":"https://schema.org","@type":"NewsArticle","generatedAt":"2026-07-28T06:20:51.496Z","headline":"Kimi 发布视觉感知基准 PerceptionBench","description":"Kimi.ai 发布 PerceptionBench，一个从当前前沿模型在 42 个基准上的失败模式中归纳出的视觉感知基准。该基准将视觉感知拆解为 10 种原子能力，并构建了 3000 道验证题，每道题只考察单一感知能力，无需推理或外部知识。","url":"https://www.aioga.com/news/cms3lggqj00c0roskhu1r5f09/","mainEntityOfPage":"https://www.aioga.com/news/cms3lggqj00c0roskhu1r5f09/","datePublished":"2026-07-27T18:45:20.000Z","dateModified":"2026-07-27T18:45:20.000Z","inLanguage":"zh-CN","publisher":{"@type":"NewsMediaOrganization","name":"Aioga","url":"https://www.aioga.com"},"citation":["https://x.com/Kimi_Moonshot/status/2081813202514681878","https://aihot.virxact.com/items/cms3lggqj00c0roskhu1r5f09"],"canonicalUrl":"https://www.aioga.com/news/cms3lggqj00c0roskhu1r5f09/","directAnswer":{"@type":"Answer","text":"Kimi.ai 发布视觉感知基准 PerceptionBench。该基准源自对当前前沿模型在 42 个基准上失败模式的归纳，包含 3000 道验证题。","url":"https://www.aioga.com/news/cms3lggqj00c0roskhu1r5f09/","dateCreated":"2026-07-27T18:45:20.000Z","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":"X：Kimi.ai (@Kimi_Moonshot) source article","url":"https://x.com/Kimi_Moonshot/status/2081813202514681878","datePublished":"2026-07-27T18:45:20.000Z","provider":{"@type":"Organization","name":"X：Kimi.ai (@Kimi_Moonshot)","url":"https://x.com/Kimi_Moonshot/status/2081813202514681878"}},{"@type":"CreativeWork","name":"AIHot archive record","url":"https://aihot.virxact.com/items/cms3lggqj00c0roskhu1r5f09","datePublished":"2026-07-27T18:45:20.000Z","provider":{"@type":"Organization","name":"AIHot","url":"https://aihot.virxact.com/items/cms3lggqj00c0roskhu1r5f09"}}],"aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","originalPublisher":{"name":"X：Kimi.ai (@Kimi_Moonshot)","url":"https://x.com/Kimi_Moonshot/status/2081813202514681878"},"article":{"id":"cms3lggqj00c0roskhu1r5f09","slug":"cms3lggqj00c0roskhu1r5f09","url":"https://www.aioga.com/news/cms3lggqj00c0roskhu1r5f09/","title":"Kimi 发布视觉感知基准 PerceptionBench","title_en":"We are releasing PerceptionBench， a benchmark that isolates visual perception and evaluates it as a …","summary":"Kimi.ai 发布 PerceptionBench，一个从当前前沿模型在 42 个基准上的失败模式中归纳出的视觉感知基准。该基准将视觉感知拆解为 10 种原子能力，并构建了 3000 道验证题，每道题只考察单一感知能力，无需推理或外部知识。","source":"X：Kimi.ai (@Kimi_Moonshot)","sourceUrl":"https://x.com/Kimi_Moonshot/status/2081813202514681878","aiHotUrl":"https://aihot.virxact.com/items/cms3lggqj00c0roskhu1r5f09","publishedAt":"2026-07-27T18:45:20.000Z","category":"产品更新","score":73,"selected":true,"articleBody":["Kimi.ai 发布 PerceptionBench，一个从当前前沿模型在 42 个基准上的失败模式中归纳出的视觉感知基准。","该基准将视觉感知拆解为 10 种原子能力，并构建了 3000 道验证题，每道题只考察单一感知能力，无需推理或外部知识。"],"articleImages":[],"mediaStatus":"none","articleBodyZh":[],"translationStatus":"","bodyOrigin":"social-summary","editorial":{"summary":"Kimi.ai 发布视觉感知基准 PerceptionBench。该基准源自对当前前沿模型在 42 个基准上失败模式的归纳，包含 3000 道验证题。","background":"PerceptionBench 将视觉感知拆解为 10 种原子能力。按照发布材料，每道验证题仅考察一种感知能力，不要求模型进行推理，也不依赖外部知识。","viewpoint":"Aioga 判断，这种单项能力拆解方式有助于更清晰地定位模型的视觉感知薄弱环节，并减少推理能力或外部知识对测试结果的干扰。","implications":"该基准可能为视觉模型评测提供更细粒度的观察框架。值得关注的是，其题目设计能否稳定区分不同模型在各类原子感知能力上的表现。","nextStep":"下一步值得关注 PerceptionBench 的具体题型、评测方法与使用方式，以及后续是否披露不同前沿模型在 10 种原子能力上的测试结果。","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":"2026-07-27T19:29:39.021Z","sourceHash":"8bf0a80bd6924531","review":{"approved":true,"groundedness":96,"clarity":94,"duplicationRisk":18,"blockingIssues":[],"notes":["“有助于更清晰地定位模型的视觉感知薄弱环节”等内容属于编辑判断，候选内容已通过“Aioga 判断”明确标注，并未冒充来源事实。","“可能为视觉模型评测提供更细粒度的观察框架”及后续关注事项属于合理展望，措辞保留了不确定性。","evidenceRefs 使用字段级引用而非具体引文；如发布格式允许，可进一步对应到支持各项事实的具体来源段落。"]},"validation":{"passed":true,"mode":"ai-auto","revisions":0,"checks":["schema","length","source-attribution","low-source-overlap","no-html","independent-ai-review"]}},"tags":["产品更新","X：Kimi.ai (@Kimi_Moonshot)"],"translations":{"zh-CN":{"title":"Kimi 发布视觉感知基准 PerceptionBench","summary":"Kimi.ai 发布 PerceptionBench，一个从当前前沿模型在 42 个基准上的失败模式中归纳出的视觉感知基准。该基准将视觉感知拆解为 10 种原子能力，并构建了 3000 道验证题，每道题只考察单一感知能力，无需推理或外部知识。","category":"产品更新","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi 发布视觉感知基准 PerceptionBench - Aioga AI资讯","description":"Kimi.ai 发布 PerceptionBench，一个从当前前沿模型在 42 个基准上的失败模式中归纳出的视觉感知基准。该基准将视觉感知拆解为 10 种原子能力，并构建了 3000 道验证题，每道题只考察单一感知能力，无需推理或外部知识。","url":"https://www.aioga.com/news/cms3lggqj00c0roskhu1r5f09/"},"en":{"title":"Kimi releases visual perception benchmark PerceptionBench","summary":"Kimi.ai released PerceptionBench, a visual perception benchmark derived from the failure patterns of current state-of-the-art models across 42 benchmarks. This benchmark breaks down visual perception into 10 atomic abilities and has created 3,000 validation questions, each testing a single perception ability, requiring no reasoning or external knowledge.","category":"Products","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi releases visual perception benchmark PerceptionBench - Aioga AI News","description":"Kimi.ai released PerceptionBench, a visual perception benchmark derived from the failure patterns of current state-of-the-art models across 42 benchmarks. This benchmark breaks dow...","url":"https://www.aioga.com/en/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:23:05.246Z"},"ja":{"title":"Kimi、視覚認識ベンチマーク PerceptionBench を発表","summary":"Kimi.ai は PerceptionBench を発表しました。これは、現在の最先端モデルの 42 のベンチマークにおける失敗パターンから抽出された視覚認知ベンチマークです。このベンチマークは視覚認知を 10 種類の基本能力に分解し、各問題が単一の認知能力のみを問う 3000 問の検証問題を構築しており、推論や外部知識は必要ありません。","category":"製品更新","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi、視覚認識ベンチマーク PerceptionBench を発表 - Aioga AIニュース","description":"Kimi.ai は PerceptionBench を発表しました。これは、現在の最先端モデルの 42 のベンチマークにおける失敗パターンから抽出された視覚認知ベンチマークです。このベンチマークは視覚認知を 10 種類の基本能力に分解し、各問題が単一の認知能力のみを問う 3000 問の検証問題を構築しており、推論や外部知識は必要ありません。","url":"https://www.aioga.com/ja/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:23:09.917Z"},"ko":{"title":"Kimi가 시각 인식 벤치마크 PerceptionBench를 발표","summary":"Kimi.ai는 현재 최첨단 모델의 42개 벤치마크에서 실패 패턴을 요약한 시각 인식 벤치마크인 PerceptionBench를 발표했습니다. 이 벤치마크는 시각 인식을 10가지 기본 능력으로 분해하고, 각 문제마다 단일 인식 능력만을 시험하는 3000개의 검증 문제를 구축했으며, 추론이나 외부 지식이 필요하지 않습니다.","category":"제품 업데이트","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi가 시각 인식 벤치마크 PerceptionBench를 발표 - Aioga AI 뉴스","description":"Kimi.ai는 현재 최첨단 모델의 42개 벤치마크에서 실패 패턴을 요약한 시각 인식 벤치마크인 PerceptionBench를 발표했습니다. 이 벤치마크는 시각 인식을 10가지 기본 능력으로 분해하고, 각 문제마다 단일 인식 능력만을 시험하는 3000개의 검증 문제를 구축했으며, 추론이나 외부 지식이 필요하지 않습니다.","url":"https://www.aioga.com/ko/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:23:57.896Z"},"es":{"title":"Kimi lanza el punto de referencia de percepción PerceptionBench","summary":"Kimi.ai lanzó PerceptionBench, un benchmark de percepción visual derivado de los modos de falla de los modelos de vanguardia actuales en 42 benchmarks. Este benchmark descompone la percepción visual en 10 capacidades atómicas y construye 3000 preguntas de verificación, cada una evaluando solo una capacidad perceptiva, sin necesidad de razonamiento ni conocimientos externos.","category":"Productos","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi lanza el punto de referencia de percepción PerceptionBench - Aioga Noticias de IA","description":"Kimi.ai lanzó PerceptionBench, un benchmark de percepción visual derivado de los modos de falla de los modelos de vanguardia actuales en 42 benchmarks. Este benchmark descompone la...","url":"https://www.aioga.com/es/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:23:50.974Z"},"fr":{"title":"Kimi publie le benchmark de perception PerceptionBench","summary":"Kimi.ai a publié PerceptionBench, un benchmark de perception visuelle dérivé des modes d'échec des modèles de pointe actuels sur 42 benchmarks. Ce benchmark décompose la perception visuelle en 10 compétences atomiques et a construit 3000 questions de validation, chaque question ne testant qu'une seule compétence de perception, sans nécessiter de raisonnement ou de connaissances externes.","category":"Produits","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi publie le benchmark de perception PerceptionBench - Aioga Actualités IA","description":"Kimi.ai a publié PerceptionBench, un benchmark de perception visuelle dérivé des modes d'échec des modèles de pointe actuels sur 42 benchmarks. Ce benchmark décompose la perception...","url":"https://www.aioga.com/fr/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:24:40.655Z"},"de":{"title":"Kimi veröffentlicht die visuelle Wahrnehmungs-Benchmark PerceptionBench","summary":"Kimi.ai hat PerceptionBench veröffentlicht, ein Benchmark für visuelle Wahrnehmung, das aus den Ausfallmustern aktueller Spitzenmodelle in 42 Benchmarks abgeleitet wurde. Dieses Benchmark zerlegt die visuelle Wahrnehmung in 10 atomare Fähigkeiten und erstellt 3000 Prüfungsaufgaben, von denen jede nur eine einzelne Wahrnehmungsfähigkeit testet, ohne dass Schlussfolgerungen oder externes Wissen erforderlich sind.","category":"产品更新","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi veröffentlicht die visuelle Wahrnehmungs-Benchmark PerceptionBench - Aioga KI-News","description":"Kimi.ai hat PerceptionBench veröffentlicht, ein Benchmark für visuelle Wahrnehmung, das aus den Ausfallmustern aktueller Spitzenmodelle in 42 Benchmarks abgeleitet wurde. Dieses Be...","url":"https://www.aioga.com/de/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:24:47.803Z"},"pt-BR":{"title":"Kimi lança o benchmark de percepção visual PerceptionBench","summary":"Kimi.ai lançou o PerceptionBench, um benchmark de percepção visual derivado dos padrões de falha de modelos de ponta atuais em 42 benchmarks. Este benchmark divide a percepção visual em 10 habilidades atômicas e construiu 3000 questões de validação, cada uma testando apenas uma única habilidade de percepção, sem necessidade de raciocínio ou conhecimento externo.","category":"产品更新","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi lança o benchmark de percepção visual PerceptionBench - Aioga Notícias de IA","description":"Kimi.ai lançou o PerceptionBench, um benchmark de percepção visual derivado dos padrões de falha de modelos de ponta atuais em 42 benchmarks. Este benchmark divide a percepção visu...","url":"https://www.aioga.com/pt-BR/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:25:24.581Z"},"ru":{"title":"Kimi выпустила визуальный эталон PerceptionBench","summary":"Kimi.ai выпустила PerceptionBench, эталон визуального восприятия, выведенный из моделей текущего передового уровня на 42 бенчмарках по их моделям ошибок. Этот эталон разбивает визуальное восприятие на 10 атомарных способностей и создает 3000 проверочных заданий, каждое из которых проверяет только одну способность восприятия, без необходимости рассуждений или внешних знаний.","category":"产品更新","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi выпустила визуальный эталон PerceptionBench - Aioga Новости ИИ","description":"Kimi.ai выпустила PerceptionBench, эталон визуального восприятия, выведенный из моделей текущего передового уровня на 42 бенчмарках по их моделям ошибок. Этот эталон разбивает визу...","url":"https://www.aioga.com/ru/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:25:30.033Z"},"ar":{"title":"Kimi أطلقت معيار الإدراك البصري PerceptionBench","summary":"أصدرت Kimi.ai PerceptionBench، وهو معيار للإدراك البصري مستخلص من أنماط الفشل للنماذج الحديثة على 42 معيارًا. يقوم هذا المعيار بتفكيك الإدراك البصري إلى 10 قدرات أساسية، وتم إنشاء 3000 سؤال للتحقق، كل سؤال يختبر قدرة إدراكية واحدة فقط، دون الحاجة إلى الاستدلال أو المعرفة الخارجية.","category":"产品更新","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi أطلقت معيار الإدراك البصري PerceptionBench - Aioga أخبار الذكاء الاصطناعي","description":"أصدرت Kimi.ai PerceptionBench، وهو معيار للإدراك البصري مستخلص من أنماط الفشل للنماذج الحديثة على 42 معيارًا. يقوم هذا المعيار بتفكيك الإدراك البصري إلى 10 قدرات أساسية، وتم إنشاء...","url":"https://www.aioga.com/ar/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:26:13.634Z"},"hi":{"title":"Kimi ने दृश्य ग्रहण मानक PerceptionBench जारी किया","summary":"Kimi.ai ने PerceptionBench जारी किया, जो वर्तमान अग्रणी मॉडलों की 42 बेंचमार्क पर विफलता पैटर्न से निष्कर्षित एक दृश्य धारणा बेंचमार्क है। यह बेंचमार्क दृश्य धारणा को 10 मूलभूत क्षमताओं में विभाजित करता है और 3000 सत्यापन प्रश्न बनाता है, प्रत्येक प्रश्न केवल एक ही धारणा क्षमता का परीक्षण करता है, जिसमें तर्क या बाहरी ज्ञान की आवश्यकता नहीं होती।","category":"产品更新","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi ने दृश्य ग्रहण मानक PerceptionBench जारी किया - Aioga AI समाचार","description":"Kimi.ai ने PerceptionBench जारी किया, जो वर्तमान अग्रणी मॉडलों की 42 बेंचमार्क पर विफलता पैटर्न से निष्कर्षित एक दृश्य धारणा बेंचमार्क है। यह बेंचमार्क दृश्य धारणा को 10 मूलभूत क्ष...","url":"https://www.aioga.com/hi/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:26:13.056Z"},"it":{"title":"Kimi lancia il benchmark di percezione PerceptionBench","summary":"Kimi.ai ha lanciato PerceptionBench, un benchmark di percezione visiva derivato dai modelli all'avanguardia attuali basato sui modelli di errore in 42 benchmark. Questo benchmark suddivide la percezione visiva in 10 abilità atomiche e ha creato 3000 quesiti di verifica, ciascuno dei quali testa una singola abilità percettiva senza necessità di ragionamento o conoscenze esterne.","category":"产品更新","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi lancia il benchmark di percezione PerceptionBench - Aioga Notizie IA","description":"Kimi.ai ha lanciato PerceptionBench, un benchmark di percezione visiva derivato dai modelli all'avanguardia attuali basato sui modelli di errore in 42 benchmark. Questo benchmark s...","url":"https://www.aioga.com/it/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:26:54.104Z"},"nl":{"title":"Kimi lanceert visuele perceptiebaseline PerceptionBench","summary":"Kimi.ai heeft PerceptionBench gelanceerd, een visueel perceptiebestand dat is afgeleid van de faalpatronen van huidige geavanceerde modellen op 42 benchmarks. Deze benchmark breekt visuele perceptie op in 10 fundamentele capaciteiten en bevat 3000 verificatievragen, waarbij elke vraag slechts één perceptievaardigheid test, zonder dat er redenen of externe kennis nodig zijn.","category":"产品更新","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi lanceert visuele perceptiebaseline PerceptionBench - Aioga AI-nieuws","description":"Kimi.ai heeft PerceptionBench gelanceerd, een visueel perceptiebestand dat is afgeleid van de faalpatronen van huidige geavanceerde modellen op 42 benchmarks. Deze benchmark breekt...","url":"https://www.aioga.com/nl/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:26:52.633Z"},"tr":{"title":"Kimi Görsel Algılama Benchmark PerceptionBench'i Yayınladı","summary":"Kimi.ai, mevcut ileri seviye modellerin 42 benchmark üzerindeki başarısızlık modellerinden türetilen bir görsel algı benchmark'ı olan PerceptionBench'i yayınladı. Bu benchmark, görsel algıyı 10 temel yeteneğe ayırıyor ve her biri yalnızca tek bir algı yeteneğini test eden, akıl yürütme veya dış bilgi gerektirmeyen 3000 doğrulama sorusu oluşturdu.","category":"产品更新","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi Görsel Algılama Benchmark PerceptionBench'i Yayınladı - Aioga AI Haberleri","description":"Kimi.ai, mevcut ileri seviye modellerin 42 benchmark üzerindeki başarısızlık modellerinden türetilen bir görsel algı benchmark'ı olan PerceptionBench'i yayınladı. Bu benchmark, gör...","url":"https://www.aioga.com/tr/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:27:34.480Z"},"vi":{"title":"Kimi phát hành chuẩn đánh giá nhận thức PerceptionBench","summary":"Kimi.ai ra mắt PerceptionBench, một chuẩn đánh giá thị giác được tổng hợp từ các mô hình tiên tiến hiện tại trên 42 tiêu chuẩn về các mô thức thất bại. Chuẩn đánh giá này chia nhận thức thị giác thành 10 khả năng cơ bản và xây dựng 3000 câu hỏi kiểm tra, mỗi câu chỉ đánh giá một khả năng nhận thức duy nhất, không cần suy luận hay kiến thức bên ngoài.","category":"产品更新","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi phát hành chuẩn đánh giá nhận thức PerceptionBench - Tin tức AI Aioga","description":"Kimi.ai ra mắt PerceptionBench, một chuẩn đánh giá thị giác được tổng hợp từ các mô hình tiên tiến hiện tại trên 42 tiêu chuẩn về các mô thức thất bại. Chuẩn đánh giá này chia nhận...","url":"https://www.aioga.com/vi/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:27:45.438Z"},"id":{"title":"Kimi merilis benchmark persepsi visual PerceptionBench","summary":"Kimi.ai merilis PerceptionBench, sebuah tolok ukur persepsi visual yang diambil dari pola kegagalan model mutakhir saat diuji pada 42 tolok ukur. Tolok ukur ini memecah persepsi visual menjadi 10 kemampuan atomik dan membangun 3000 soal verifikasi, di mana setiap soal hanya menguji satu kemampuan persepsi, tanpa memerlukan penalaran atau pengetahuan eksternal.","category":"产品更新","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi merilis benchmark persepsi visual PerceptionBench - Berita AI Aioga","description":"Kimi.ai merilis PerceptionBench, sebuah tolok ukur persepsi visual yang diambil dari pola kegagalan model mutakhir saat diuji pada 42 tolok ukur. Tolok ukur ini memecah persepsi vi...","url":"https://www.aioga.com/id/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:28:19.601Z"},"th":{"title":"Kimi เปิดตัวมาตรฐานการรับรู้ด้วยสายตา PerceptionBench","summary":"Kimi.ai เปิดตัว PerceptionBench ซึ่งเป็นเกณฑ์การรับรู้ทางสายตาที่สรุปจากรูปแบบความล้มเหลวของโมเดลล้ำสมัยปัจจุบันบนเกณฑ์มาตรฐาน 42 ตัว เกณฑ์นี้แบ่งการรับรู้ทางสายตาออกเป็น 10 ความสามารถพื้นฐาน และสร้างข้อสอบตรวจสอบ 3,000 ข้อ แต่ละข้อทดสอบเพียงความสามารถการรับรู้เดียวโดยไม่จำเป็นต้องใช้การให้เหตุผลหรือความรู้ภายนอก","category":"产品更新","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi เปิดตัวมาตรฐานการรับรู้ด้วยสายตา PerceptionBench - ข่าว AI Aioga","description":"Kimi.ai เปิดตัว PerceptionBench ซึ่งเป็นเกณฑ์การรับรู้ทางสายตาที่สรุปจากรูปแบบความล้มเหลวของโมเดลล้ำสมัยปัจจุบันบนเกณฑ์มาตรฐาน 42 ตัว เกณฑ์นี้แบ่งการรับรู้ทางสายตาออกเป็น 10 ความสา...","url":"https://www.aioga.com/th/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:28:33.772Z"},"pl":{"title":"Kimi wydaje standard PerceptionBench do percepcji wizualnej","summary":"Kimi.ai wydało PerceptionBench, benchmark percepcji wizualnej opracowany na podstawie wzorców błędów obecnych w najbardziej zaawansowanych modelach w 42 benchmarkach. Benchmark ten rozkłada percepcję wizualną na 10 podstawowych umiejętności i stworzył 3000 zadań weryfikacyjnych, z których każde sprawdza tylko jedną umiejętność percepcyjną, nie wymagając rozumowania ani wiedzy zewnętrznej.","category":"产品更新","source":"X：Kimi.ai (@Kimi_Moonshot)","aggregationSource":"X：Kimi.ai (@Kimi_Moonshot)","pageTitle":"Kimi wydaje standard PerceptionBench do percepcji wizualnej - Aioga Wiadomości AI","description":"Kimi.ai wydało PerceptionBench, benchmark percepcji wizualnej opracowany na podstawie wzorców błędów obecnych w najbardziej zaawansowanych modelach w 42 benchmarkach. Benchmark ten...","url":"https://www.aioga.com/pl/news/cms3lggqj00c0roskhu1r5f09/","contentTranslated":true,"sourceHash":"bf1d78a3c1fc15c9","translatedAt":"2026-07-27T19:29:11.568Z"}}}}