@lobehub/chat 1.128.1 → 1.128.2

package/CHANGELOG.md

@@ -2,6 +2,32 @@
 
 # Changelog
 
+### [Version 1.128.2](https://github.com/lobehub/lobe-chat/compare/v1.128.1...v1.128.2)
+
+<sup>Released on **2025-09-13**</sup>
+
+#### 💄 Styles
+
+- **misc**: Update i18n, Update model configs.
+
+<br/>
+
+<details>
+<summary><kbd>Improvements and Fixes</kbd></summary>
+
+#### Styles
+
+- **misc**: Update i18n, closes [#9237](https://github.com/lobehub/lobe-chat/issues/9237) ([642dc3b](https://github.com/lobehub/lobe-chat/commit/642dc3b))
+- **misc**: Update model configs, closes [#9170](https://github.com/lobehub/lobe-chat/issues/9170) ([f89b730](https://github.com/lobehub/lobe-chat/commit/f89b730))
+
+</details>
+
+<div align="right">
+
+[![](https://img.shields.io/badge/-BACK_TO_TOP-151515?style=flat-square)](#readme-top)
+
+</div>
+
 ### [Version 1.128.1](https://github.com/lobehub/lobe-chat/compare/v1.128.0...v1.128.1)
 
 <sup>Released on **2025-09-12**</sup>

package/changelog/v1.json

@@ -1,4 +1,13 @@
 [
+  {
+    "children": {
+      "improvements": [
+        "Update i18n, Update model configs."
+      ]
+    },
+    "date": "2025-09-13",
+    "version": "1.128.2"
+  },
   {
     "children": {
       "improvements": [

package/locales/ar/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "نسخة المعاينة لنموذج Qwen 3 Max من سلسلة Tongyi Qianwen، مع تحسينات كبيرة في القدرات العامة مقارنة بسلسلة 2.5، بما في ذلك فهم النصوص العامة باللغتين الصينية والإنجليزية، الالتزام بالتعليمات المعقدة، المهام المفتوحة الذاتية، القدرات متعددة اللغات، واستدعاء الأدوات؛ مع تقليل الأوهام المعرفية للنموذج."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "نموذج مفتوح المصدر من الجيل الجديد لوضع عدم التفكير مبني على Qwen3، يتميز بفهم أفضل للنصوص الصينية مقارنة بالإصدار السابق (Tongyi Qianwen 3-235B-A22B-Instruct-2507)، مع تعزيز في قدرات الاستدلال المنطقي وأداء أفضل في مهام توليد النصوص."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "نموذج مفتوح المصدر من الجيل الجديد لوضع التفكير مبني على Qwen3، يتميز بتحسين في الالتزام بالتعليمات مقارنة بالإصدار السابق (Tongyi Qianwen 3-235B-A22B-Thinking-2507)، مع ردود ملخصة وأكثر إيجازًا من النموذج."
+  },
   "qwq": {
     "description": "QwQ هو نموذج بحث تجريبي يركز على تحسين قدرات الاستدلال للذكاء الاصطناعي."
   },

package/locales/bg-BG/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "Preview версия на Max модела от серията Qwen 3, с голямо подобрение спрямо серия 2.5 в общите универсални способности, разбиране на текст на китайски и английски, следване на сложни инструкции, субективни отворени задачи, многоезични способности и използване на инструменти; моделът има по-малко халюцинации на знания."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "Базирано на Qwen3, ново поколение отворен модел без мисловен режим, който предлага по-добро разбиране на китайски текстове, подобрени логически умения и по-добри резултати при задачи за генериране на текст в сравнение с предишната версия (Tongyi Qianwen 3-235B-A22B-Instruct-2507)."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "Базирано на Qwen3, ново поколение отворен модел с мисловен режим, който подобрява спазването на инструкции и предоставя по-кратки и точни обобщения в сравнение с предишната версия (Tongyi Qianwen 3-235B-A22B-Thinking-2507)."
+  },
   "qwq": {
     "description": "QwQ е експериментален изследователски модел, който се фокусира върху подобряване на AI разсъдъчните способности."
   },

package/locales/de-DE/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "Die Preview-Version des Max-Modells der Tongyi Qianwen 3-Serie zeigt im Vergleich zur 2.5-Serie eine deutliche Steigerung der allgemeinen Fähigkeiten, einschließlich verbesserter chinesisch- und englischsprachiger Textverständnisfähigkeiten, komplexer Befolgung von Anweisungen, subjektiver offener Aufgaben, Mehrsprachigkeit und Werkzeugaufruf-Fähigkeiten; das Modell zeigt weniger Wissenshalluzinationen."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "Ein neues Open-Source-Modell der nächsten Generation im Nicht-Denk-Modus basierend auf Qwen3. Im Vergleich zur vorherigen Version (Tongyi Qianwen 3-235B-A22B-Instruct-2507) bietet es eine verbesserte chinesische Textverständnisfähigkeit, verstärkte logische Schlussfolgerungen und bessere Leistung bei textgenerierenden Aufgaben."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "Ein neues Open-Source-Modell der nächsten Generation im Denkmodus basierend auf Qwen3. Im Vergleich zur vorherigen Version (Tongyi Qianwen 3-235B-A22B-Thinking-2507) wurde die Befehlsbefolgung verbessert und die Modellantworten sind prägnanter zusammengefasst."
+  },
   "qwq": {
     "description": "QwQ ist ein experimentelles Forschungsmodell, das sich auf die Verbesserung der KI-Inferenzfähigkeiten konzentriert."
   },

package/locales/en-US/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "The Qwen 3 series Max model preview version shows a significant overall improvement compared to the 2.5 series in general capabilities, including Chinese and English text understanding, complex instruction adherence, subjective open tasks, multilingual capabilities, and tool invocation. The model also exhibits fewer knowledge hallucinations."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "A new generation of non-thinking mode open-source model based on Qwen3. Compared to the previous version (Tongyi Qianwen 3-235B-A22B-Instruct-2507), it offers better Chinese text comprehension, enhanced logical reasoning abilities, and improved performance in text generation tasks."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "A new generation of thinking mode open-source model based on Qwen3. Compared to the previous version (Tongyi Qianwen 3-235B-A22B-Thinking-2507), it features improved instruction-following capabilities and more concise model-generated summaries."
+  },
   "qwq": {
     "description": "QwQ is an experimental research model focused on improving AI reasoning capabilities."
   },

package/locales/es-ES/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "Versión preliminar del modelo Max de la serie Qwen 3, que presenta una mejora significativa en la capacidad general en comparación con la serie 2.5, incluyendo comprensión de texto en chino e inglés, cumplimiento de instrucciones complejas, tareas abiertas subjetivas, capacidades multilingües y llamadas a herramientas; además, reduce notablemente las alucinaciones de conocimiento del modelo."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "Modelo de código abierto de nueva generación basado en Qwen3 en modo no reflexivo, que ofrece una mejor comprensión del texto en chino, mayor capacidad de razonamiento lógico y un mejor desempeño en tareas de generación de texto en comparación con la versión anterior (Tongyi Qianwen 3-235B-A22B-Instruct-2507)."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "Modelo de código abierto de nueva generación basado en Qwen3 en modo reflexivo, que mejora la capacidad de seguir instrucciones y ofrece respuestas más concisas en comparación con la versión anterior (Tongyi Qianwen 3-235B-A22B-Thinking-2507)."
+  },
   "qwq": {
     "description": "QwQ es un modelo de investigación experimental que se centra en mejorar la capacidad de razonamiento de la IA."
   },

package/locales/fa-IR/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "نسخه پیش‌نمایش مدل Max از سری Qwen 3، نسبت به سری 2.5 بهبود قابل توجهی در توانایی‌های عمومی، درک متن‌های چندزبانه چینی و انگلیسی، پیروی از دستورات پیچیده، انجام وظایف باز و ذهنی، پشتیبانی چندزبانه و فراخوانی ابزارها دارد؛ همچنین خطاهای دانش مدل کاهش یافته است."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "مدل متن‌باز نسل جدید بدون حالت تفکر مبتنی بر Qwen3، که نسبت به نسخه قبلی (Tongyi Qianwen 3-235B-A22B-Instruct-2507) در درک متن‌های چینی بهتر است، توانایی استدلال منطقی بهبود یافته و عملکرد بهتری در وظایف تولید متن دارد."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "مدل متن‌باز نسل جدید با حالت تفکر مبتنی بر Qwen3، که نسبت به نسخه قبلی (Tongyi Qianwen 3-235B-A22B-Thinking-2507) در پیروی از دستورات پیشرفت داشته و پاسخ‌های مدل خلاصه‌تر شده‌اند."
+  },
   "qwq": {
     "description": "QwQ یک مدل تحقیقاتی تجربی است که بر بهبود توانایی استدلال AI تمرکز دارد."
   },

package/locales/fr-FR/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "Version Preview du modèle Max de la série Qwen 3, avec une amélioration significative des capacités générales par rapport à la série 2.5, notamment en compréhension de texte général bilingue (chinois et anglais), respect des instructions complexes, tâches ouvertes subjectives, capacités multilingues et appels d'outils ; le modèle présente également moins d'hallucinations de connaissances."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "Modèle open source de nouvelle génération en mode non réflexif basé sur Qwen3, offrant une meilleure compréhension du texte en chinois, des capacités de raisonnement logique renforcées et de meilleures performances dans les tâches de génération de texte par rapport à la version précédente (Tongyi Qianwen 3-235B-A22B-Instruct-2507)."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "Modèle open source de nouvelle génération en mode réflexif basé sur Qwen3, avec une meilleure conformité aux instructions et des réponses plus concises dans les résumés par rapport à la version précédente (Tongyi Qianwen 3-235B-A22B-Thinking-2507)."
+  },
   "qwq": {
     "description": "QwQ est un modèle de recherche expérimental, axé sur l'amélioration des capacités de raisonnement de l'IA."
   },

package/locales/it-IT/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "Versione preview del modello Max della serie Qwen 3, che presenta un miglioramento significativo rispetto alla serie 2.5 nelle capacità generali, comprensione del testo in cinese e inglese, rispetto di istruzioni complesse, compiti soggettivi aperti, capacità multilingue e chiamata di strumenti; il modello presenta meno allucinazioni di conoscenza."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "Modello open source di nuova generazione basato su Qwen3 in modalità non riflessiva, con una migliore comprensione del testo in cinese rispetto alla versione precedente (Tongyi Qianwen 3-235B-A22B-Instruct-2507), capacità di ragionamento logico potenziate e prestazioni superiori nelle attività di generazione di testo."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "Modello open source di nuova generazione basato su Qwen3 in modalità riflessiva, con migliorata capacità di seguire le istruzioni rispetto alla versione precedente (Tongyi Qianwen 3-235B-A22B-Thinking-2507) e risposte di sintesi più concise."
+  },
   "qwq": {
     "description": "QwQ è un modello di ricerca sperimentale, focalizzato sul miglioramento delle capacità di ragionamento dell'IA."
   },

package/locales/ja-JP/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "通義千問3シリーズMaxモデルのプレビュー版で、2.5シリーズに比べて全体的な汎用能力が大幅に向上し、中日両言語の汎用テキスト理解能力、複雑な指示遵守能力、主観的なオープンタスク能力、多言語能力、ツール呼び出し能力が著しく強化されました。モデルの知識幻覚も減少しています。"
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "Qwen3に基づく次世代の非思考モードのオープンソースモデルで、前バージョン（通義千問3-235B-A22B-Instruct-2507）と比べて中国語テキストの理解能力が向上し、論理推論能力が強化され、テキスト生成タスクのパフォーマンスがより優れています。"
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "Qwen3に基づく次世代の思考モードのオープンソースモデルで、前バージョン（通義千問3-235B-A22B-Thinking-2507）と比べて指示遵守能力が向上し、モデルの要約応答がより簡潔になっています。"
+  },
   "qwq": {
     "description": "QwQはAIの推論能力を向上させることに特化した実験的研究モデルです。"
   },

package/locales/ko-KR/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "통의천문3 시리즈 Max 모델 프리뷰 버전으로, 2.5 시리즈에 비해 전반적인 범용 능력이 크게 향상되었으며, 중영문 범용 텍스트 이해 능력, 복잡한 지시 준수 능력, 주관적 개방형 작업 능력, 다국어 능력, 도구 호출 능력이 모두 크게 강화되었습니다. 모델의 지식 환각도 감소하였습니다."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "Qwen3 기반의 차세대 비사고 모드 오픈 소스 모델로, 이전 버전(통의천문3-235B-A22B-Instruct-2507)과 비교하여 중국어 텍스트 이해 능력이 향상되었고, 논리 추론 능력이 강화되었으며, 텍스트 생성 작업에서 더 우수한 성능을 보입니다."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "Qwen3 기반의 차세대 사고 모드 오픈 소스 모델로, 이전 버전(통의천문3-235B-A22B-Thinking-2507)과 비교하여 명령 준수 능력이 향상되었고, 모델의 요약 응답이 더욱 간결해졌습니다."
+  },
   "qwq": {
     "description": "QwQ는 AI 추론 능력을 향상시키는 데 중점을 둔 실험 연구 모델입니다."
   },

package/locales/nl-NL/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "Previewversie van het Qwen 3 Max-model uit de Tongyi Qianwen 3-serie, met aanzienlijke verbeteringen ten opzichte van de 2.5-serie in algemene capaciteiten, tweetalig tekstbegrip (Chinees en Engels), complexe instructienaleving, subjectieve open taken, meertalige vaardigheden en toolaanroepen; het model vertoont minder kennisillusies."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "Een nieuwe generatie open-source model zonder denkmodus gebaseerd op Qwen3, met verbeterde Chinese tekstbegrip, versterkte logische redeneervaardigheden en betere prestaties bij tekstgeneratietaken vergeleken met de vorige versie (Tongyi Qianwen 3-235B-A22B-Instruct-2507)."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "Een nieuwe generatie open-source model met denkmodus gebaseerd op Qwen3, met verbeterde instructienaleving en meer beknopte samenvattende antwoorden vergeleken met de vorige versie (Tongyi Qianwen 3-235B-A22B-Thinking-2507)."
+  },
   "qwq": {
     "description": "QwQ is een experimenteel onderzoeksmodel dat zich richt op het verbeteren van de AI-redeneringscapaciteiten."
   },

package/locales/pl-PL/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "Preview modelu Max z serii Qwen 3, który w porównaniu do serii 2.5 znacząco poprawia ogólne zdolności, w tym rozumienie tekstu w języku chińskim i angielskim, przestrzeganie złożonych instrukcji, zdolności do zadań otwartych i subiektywnych, wielojęzyczność oraz wywoływanie narzędzi; model generuje mniej halucynacji wiedzy."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "Nowa generacja otwartego modelu bez trybu myślenia oparta na Qwen3, która w porównaniu z poprzednią wersją (Tongyi Qianwen 3-235B-A22B-Instruct-2507) cechuje się lepszym rozumieniem tekstu w języku chińskim, wzmocnionymi zdolnościami wnioskowania logicznego oraz lepszą wydajnością w zadaniach generowania tekstu."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "Nowa generacja otwartego modelu z trybem myślenia oparta na Qwen3, która w porównaniu z poprzednią wersją (Tongyi Qianwen 3-235B-A22B-Thinking-2507) wykazuje poprawę w przestrzeganiu instrukcji oraz bardziej zwięzłe podsumowania w odpowiedziach modelu."
+  },
   "qwq": {
     "description": "QwQ to eksperymentalny model badawczy, skoncentrowany na zwiększeniu zdolności wnioskowania AI."
   },

package/locales/pt-BR/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "Versão Preview do modelo Max da série Qwen 3, com melhorias significativas em relação à série 2.5 em capacidades gerais, compreensão de texto em chinês e inglês, conformidade com instruções complexas, tarefas subjetivas abertas, multilinguismo e chamadas de ferramentas; o modelo apresenta menos alucinações de conhecimento."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "Modelo open source de nova geração no modo não reflexivo baseado no Qwen3, que apresenta melhor compreensão de texto em chinês, capacidades aprimoradas de raciocínio lógico e desempenho superior em tarefas de geração de texto em comparação com a versão anterior (Tongyi Qianwen 3-235B-A22B-Instruct-2507)."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "Modelo open source de nova geração no modo reflexivo baseado no Qwen3, que oferece melhor conformidade com instruções e respostas mais concisas em resumos, em comparação com a versão anterior (Tongyi Qianwen 3-235B-A22B-Thinking-2507)."
+  },
   "qwq": {
     "description": "QwQ é um modelo de pesquisa experimental, focado em melhorar a capacidade de raciocínio da IA."
   },

package/locales/ru-RU/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "Предварительная версия модели серии Qwen 3 Max, которая значительно превосходит серию 2.5 по универсальным возможностям, включая понимание текста на китайском и английском языках, выполнение сложных инструкций, способности к субъективным открытым задачам, многоязычность и вызов инструментов; модель демонстрирует меньше искажений знаний."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "Новая генерация открытой модели без режима мышления на базе Qwen3, которая по сравнению с предыдущей версией (通义千问3-235B-A22B-Instruct-2507) обладает улучшенными способностями понимания китайского текста, усиленными логическими рассуждениями и лучшими результатами в задачах генерации текста."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "Новая генерация открытой модели с режимом мышления на базе Qwen3, которая по сравнению с предыдущей версией (通义千问3-235B-A22B-Thinking-2507) демонстрирует улучшенное следование инструкциям и более лаконичные ответы модели."
+  },
   "qwq": {
     "description": "QwQ — это экспериментальная исследовательская модель, сосредоточенная на повышении возможностей вывода ИИ."
   },

package/locales/tr-TR/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "Tongyi Qianwen 3 serisi Max modelinin önizleme sürümüdür. 2.5 serisine kıyasla genel yeteneklerde büyük gelişmeler göstermiştir; Çince ve İngilizce genel metin anlama, karmaşık talimat uyumu, öznel açık görevler, çok dilli yetenekler ve araç çağrısı yetenekleri belirgin şekilde artmıştır; model bilgi yanılgıları daha azdır."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "Qwen3 tabanlı yeni nesil düşünmeden çalışan açık kaynak modeli, önceki sürüme (Tongyi Qianwen 3-235B-A22B-Instruct-2507) kıyasla Çince metin anlama yeteneği daha iyi, mantıksal çıkarım yeteneği geliştirilmiş ve metin üretimi görevlerinde daha başarılıdır."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "Qwen3 tabanlı yeni nesil düşünme modlu açık kaynak modeli, önceki sürüme (Tongyi Qianwen 3-235B-A22B-Thinking-2507) kıyasla komutlara uyum yeteneği artırılmış ve modelin özetleyici yanıtları daha özlü hale getirilmiştir."
+  },
   "qwq": {
     "description": "QwQ, AI akıl yürütme yeteneklerini artırmaya odaklanan deneysel bir araştırma modelidir."
   },

package/locales/vi-VN/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "Phiên bản xem trước mô hình Max của dòng Qwen 3, so với dòng 2.5 có sự cải thiện lớn về năng lực tổng quát, khả năng hiểu văn bản song ngữ Trung-Anh, tuân thủ chỉ thị phức tạp, thực hiện nhiệm vụ mở chủ quan, đa ngôn ngữ và gọi công cụ đều được tăng cường rõ rệt; mô hình giảm thiểu ảo giác kiến thức."
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "Mô hình mã nguồn mở thế hệ mới không có chế độ suy nghĩ dựa trên Qwen3, so với phiên bản trước (Thông Nghĩa Thiên Vấn 3-235B-A22B-Instruct-2507) có khả năng hiểu văn bản tiếng Trung tốt hơn, năng lực suy luận logic được cải thiện, và hiệu suất trong các nhiệm vụ tạo văn bản cũng tốt hơn."
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "Mô hình mã nguồn mở thế hệ mới có chế độ suy nghĩ dựa trên Qwen3, so với phiên bản trước (Thông Nghĩa Thiên Vấn 3-235B-A22B-Thinking-2507) có khả năng tuân thủ chỉ dẫn được nâng cao, và các phản hồi tóm tắt của mô hình trở nên ngắn gọn hơn."
+  },
   "qwq": {
     "description": "QwQ là một mô hình nghiên cứu thử nghiệm, tập trung vào việc nâng cao khả năng suy luận của AI."
   },

package/locales/zh-CN/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "通义千问3系列Max模型Preview版本，相较2.5系列整体通用能力有大幅度提升，中英文通用文本理解能力、复杂指令遵循能力、主观开放任务能力、多语言能力、工具调用能力均显著增强；模型知识幻觉更少。"
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "基于 Qwen3 的新一代非思考模式开源模型，相较上一版本（通义千问3-235B-A22B-Instruct-2507）中文文本理解能力更佳、逻辑推理能力有增强、文本生成类任务表现更好。"
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "基于 Qwen3 的新一代思考模式开源模型，相较上一版本（通义千问3-235B-A22B-Thinking-2507）指令遵循能力有提升、模型总结回复更加精简。"
+  },
   "qwq": {
     "description": "QwQ 是 Qwen 系列的推理模型。与传统的指令调优模型相比，QwQ 具备思考和推理的能力，能够在下游任务中，尤其是困难问题上，显著提升性能。QwQ-32B 是中型推理模型，能够在与最先进的推理模型（如 DeepSeek-R1、o1-mini）竞争时取得可观的表现。"
   },

package/locales/zh-TW/models.json

@@ -2615,6 +2615,12 @@
   "qwen3-max-preview": {
     "description": "通義千問3系列 Max 模型 Preview 版本，相較 2.5 系列整體通用能力有大幅度提升，中英文通用文本理解能力、複雜指令遵循能力、主觀開放任務能力、多語言能力、工具調用能力均顯著增強；模型知識幻覺更少。"
   },
+  "qwen3-next-80b-a3b-instruct": {
+    "description": "基於 Qwen3 的新一代非思考模式開源模型，相較上一版本（通義千問3-235B-A22B-Instruct-2507）中文文本理解能力更佳、邏輯推理能力有增強、文本生成類任務表現更好。"
+  },
+  "qwen3-next-80b-a3b-thinking": {
+    "description": "基於 Qwen3 的新一代思考模式開源模型，相較上一版本（通義千問3-235B-A22B-Thinking-2507）指令遵循能力有提升、模型總結回覆更加精簡。"
+  },
   "qwq": {
     "description": "QwQ 是一個實驗研究模型，專注於提高 AI 推理能力。"
   },

package/package.json

@@ -1,6 +1,6 @@
 {
   "name": "@lobehub/chat",
-  "version": "1.128.1",
+  "version": "1.128.2",
   "description": "Lobe Chat - an open-source, high-performance chatbot framework that supports speech synthesis, multimodal, and extensible Function Call plugin system. Supports one-click free deployment of your private ChatGPT/LLM web application.",
   "keywords": [
     "framework",

package/packages/model-bank/src/aiModels/deepseek.ts

@@ -17,8 +17,8 @@ const deepseekChatModels: AIChatModelCard[] = [
       currency: 'CNY',
       units: [
         { name: 'textInput_cacheRead', rate: 0.5, strategy: 'fixed', unit: 'millionTokens' },
-        { name: 'textInput', rate: 2, strategy: 'fixed', unit: 'millionTokens' },
-        { name: 'textOutput', rate: 8, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textInput', rate: 4, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textOutput', rate: 12, strategy: 'fixed', unit: 'millionTokens' },
       ],
     },
     releasedAt: '2025-08-21',
@@ -39,9 +39,9 @@ const deepseekChatModels: AIChatModelCard[] = [
     pricing: {
       currency: 'CNY',
       units: [
-        { name: 'textInput_cacheRead', rate: 1, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textInput_cacheRead', rate: 0.5, strategy: 'fixed', unit: 'millionTokens' },
         { name: 'textInput', rate: 4, strategy: 'fixed', unit: 'millionTokens' },
-        { name: 'textOutput', rate: 16, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textOutput', rate: 12, strategy: 'fixed', unit: 'millionTokens' },
       ],
     },
     releasedAt: '2025-08-21',

package/packages/model-bank/src/aiModels/google.ts

@@ -18,7 +18,15 @@ const googleChatModels: AIChatModelCard[] = [
     maxOutput: 65_536,
     pricing: {
       units: [
-        { name: 'textInput_cacheRead', rate: 0.31, strategy: 'fixed', unit: 'millionTokens' },
+        {
+          name: 'textInput_cacheRead',
+          strategy: 'tiered',
+          tiers: [
+            { rate: 0.31, upTo: 200_000 },
+            { rate: 0.625, upTo: 'infinity' },
+          ],
+          unit: 'millionTokens',
+        },
         {
           name: 'textInput',
           strategy: 'tiered',
@@ -57,12 +65,20 @@ const googleChatModels: AIChatModelCard[] = [
     contextWindowTokens: 1_048_576 + 65_536,
     description:
       'Gemini 2.5 Pro Preview 是 Google 最先进的思维模型，能够对代码、数学和STEM领域的复杂问题进行推理，以及使用长上下文分析大型数据集、代码库和文档。',
-    displayName: 'Gemini 2.5 Pro Preview 06-05 (Paid)',
+    displayName: 'Gemini 2.5 Pro Preview 06-05',
     id: 'gemini-2.5-pro-preview-06-05',
     maxOutput: 65_536,
     pricing: {
       units: [
-        { name: 'textInput_cacheRead', rate: 0.31, strategy: 'fixed', unit: 'millionTokens' },
+        {
+          name: 'textInput_cacheRead',
+          strategy: 'tiered',
+          tiers: [
+            { rate: 0.31, upTo: 200_000 },
+            { rate: 0.625, upTo: 'infinity' },
+          ],
+          unit: 'millionTokens',
+        },
         {
           name: 'textInput',
           strategy: 'tiered',
@@ -101,12 +117,20 @@ const googleChatModels: AIChatModelCard[] = [
     contextWindowTokens: 1_048_576 + 65_536,
     description:
       'Gemini 2.5 Pro Preview 是 Google 最先进的思维模型，能够对代码、数学和STEM领域的复杂问题进行推理，以及使用长上下文分析大型数据集、代码库和文档。',
-    displayName: 'Gemini 2.5 Pro Preview 05-06 (Paid)',
+    displayName: 'Gemini 2.5 Pro Preview 05-06',
     id: 'gemini-2.5-pro-preview-05-06',
     maxOutput: 65_536,
     pricing: {
       units: [
-        { name: 'textInput_cacheRead', rate: 0.31, strategy: 'fixed', unit: 'millionTokens' },
+        {
+          name: 'textInput_cacheRead',
+          strategy: 'tiered',
+          tiers: [
+            { rate: 0.31, upTo: 200_000 },
+            { rate: 0.625, upTo: 'infinity' },
+          ],
+          unit: 'millionTokens',
+        },
         {
           name: 'textInput',
           strategy: 'tiered',
@@ -479,7 +503,7 @@ const googleChatModels: AIChatModelCard[] = [
     contextWindowTokens: 2_008_192,
     description:
       'Gemini 1.5 Pro 002 是最新的生产就绪模型，提供更高质量的输出，特别在数学、长上下文和视觉任务方面有显著提升。',
-    displayName: 'Gemini 1.5 Pro 002 (Paid)',
+    displayName: 'Gemini 1.5 Pro 002',
     id: 'gemini-1.5-pro-002', // Deprecated on 2025-09-24
     maxOutput: 8192,
     pricing: {

package/packages/model-bank/src/aiModels/groq.ts

@@ -73,24 +73,6 @@ const groqChatModels: AIChatModelCard[] = [
     displayName: 'Kimi K2 0905',
     enabled: true,
     id: 'moonshotai/kimi-k2-instruct-0905',
-    pricing: {
-      units: [
-        { name: 'textInput', rate: 1, strategy: 'fixed', unit: 'millionTokens' },
-        { name: 'textOutput', rate: 3, strategy: 'fixed', unit: 'millionTokens' },
-      ],
-    },
-    releasedAt: '2025-09-05',
-    type: 'chat',
-  },
-  {
-    abilities: {
-      functionCall: true,
-    },
-    contextWindowTokens: 131_072,
-    description:
-      'kimi-k2 是一款具备超强代码和 Agent 能力的 MoE 架构基础模型，总参数 1T，激活参数 32B。在通用知识推理、编程、数学、Agent 等主要类别的基准性能测试中，K2 模型的性能超过其他主流开源模型。',
-    displayName: 'Kimi K2 0711',
-    id: 'moonshotai/kimi-k2-instruct',
     maxOutput: 16_384,
     pricing: {
       units: [
@@ -98,7 +80,7 @@ const groqChatModels: AIChatModelCard[] = [
         { name: 'textOutput', rate: 3, strategy: 'fixed', unit: 'millionTokens' },
       ],
     },
-    releasedAt: '2025-07-11',
+    releasedAt: '2025-09-05',
     type: 'chat',
   },
   {

package/packages/model-bank/src/aiModels/modelscope.ts

@@ -4,14 +4,20 @@ const modelscopeChatModels: AIChatModelCard[] = [
   {
     abilities: {
       functionCall: true,
+      reasoning: true,
     },
-    contextWindowTokens: 262_144,
-    description:
-      'kimi-k2-0905-preview 模型上下文长度为 256k，具备更强的 Agentic Coding 能力、更突出的前端代码的美观度和实用性、以及更好的上下文理解能力。',
-    displayName: 'Kimi K2 0905',
-    enabled: true,
-    id: 'moonshotai/Kimi-K2-Instruct-0905',
-    releasedAt: '2025-09-05',
+    contextWindowTokens: 131_072,
+    displayName: 'Qwen3 Next 80B A3B Thinking',
+    id: 'Qwen/Qwen3-Next-80B-A3B-Thinking',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+    },
+    contextWindowTokens: 131_072,
+    displayName: 'Qwen3 Next 80B A3B Instruct',
+    id: 'Qwen/Qwen3-Next-80B-A3B-Instruct',
     type: 'chat',
   },
   {
@@ -20,8 +26,8 @@ const modelscopeChatModels: AIChatModelCard[] = [
       reasoning: true,
     },
     contextWindowTokens: 131_072,
-    description: 'DeepSeek-V3.1 模型为混合推理架构模型，同时支持思考模式与非思考模式。',
-    displayName: 'DeepSeek-V3.1',
+    description: 'DeepSeek V3.1 模型为混合推理架构模型，同时支持思考模式与非思考模式。',
+    displayName: 'DeepSeek V3.1',
     enabled: true,
     id: 'deepseek-ai/DeepSeek-V3.1',
     type: 'chat',
@@ -34,7 +40,7 @@ const modelscopeChatModels: AIChatModelCard[] = [
     contextWindowTokens: 131_072,
     description:
       'DeepSeek R1 通过利用增加的计算资源和在后训练过程中引入算法优化机制，显著提高了其推理和推断能力的深度。该模型在各种基准评估中表现出色，包括数学、编程和一般逻辑方面。其整体性能现已接近领先模型，如 O3 和 Gemini 2.5 Pro。',
-    displayName: 'DeepSeek-R1-0528',
+    displayName: 'DeepSeek R1 0528',
     id: 'deepseek-ai/DeepSeek-R1-0528',
     type: 'chat',
   },
@@ -44,8 +50,8 @@ const modelscopeChatModels: AIChatModelCard[] = [
       reasoning: true,
     },
     contextWindowTokens: 131_072,
-    description: 'DeepSeek-R1是DeepSeek最新的推理模型，专注于复杂推理任务。',
-    displayName: 'DeepSeek-R1',
+    description: 'DeepSeek R1是DeepSeek最新的推理模型，专注于复杂推理任务。',
+    displayName: 'DeepSeek R1',
     id: 'deepseek-ai/DeepSeek-R1',
     type: 'chat',
   },
@@ -54,8 +60,8 @@ const modelscopeChatModels: AIChatModelCard[] = [
       functionCall: true,
     },
     contextWindowTokens: 131_072,
-    description: 'DeepSeek-V3是DeepSeek第三代模型的最新版本，具有强大的推理和对话能力。',
-    displayName: 'DeepSeek-V3',
+    description: 'DeepSeek V3是DeepSeek第三代模型的最新版本，具有强大的推理和对话能力。',
+    displayName: 'DeepSeek V3',
     id: 'deepseek-ai/DeepSeek-V3',
     type: 'chat',
   },
@@ -64,8 +70,8 @@ const modelscopeChatModels: AIChatModelCard[] = [
       functionCall: true,
     },
     contextWindowTokens: 131_072,
-    description: 'Qwen3-235B-A22B是通义千问3代超大规模模型，提供顶级的AI能力。',
-    displayName: 'Qwen3-235B-A22B',
+    description: 'Qwen3 235B A22B是通义千问3代超大规模模型，提供顶级的AI能力。',
+    displayName: 'Qwen3 235B A22B',
     id: 'Qwen/Qwen3-235B-A22B',
     type: 'chat',
   },
@@ -74,8 +80,8 @@ const modelscopeChatModels: AIChatModelCard[] = [
       functionCall: true,
     },
     contextWindowTokens: 131_072,
-    description: 'Qwen3-32B是通义千问3代模型，具有强大的推理和对话能力。',
-    displayName: 'Qwen3-32B',
+    description: 'Qwen3 32B是通义千问3代模型，具有强大的推理和对话能力。',
+    displayName: 'Qwen3 32B',
     id: 'Qwen/Qwen3-32B',
     type: 'chat',
   },

package/packages/model-bank/src/aiModels/novita.ts

@@ -2,6 +2,52 @@ import { AIChatModelCard } from '../types/aiModel';
 
 // https://novita.ai/pricing
 const novitaChatModels: AIChatModelCard[] = [
+  {
+    abilities: {
+      functionCall: true,
+      reasoning: true,
+    },
+    contextWindowTokens: 65_536,
+    displayName: 'Qwen3 Next 80B A3B Thinking',
+    id: 'qwen/qwen3-next-80b-a3b-thinking',
+    maxOutput: 65_536,
+    pricing: {
+      units: [
+        { name: 'textInput', rate: 0.15, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textOutput', rate: 1.5, strategy: 'fixed', unit: 'millionTokens' },
+      ],
+    },
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+    },
+    contextWindowTokens: 65_536,
+    displayName: 'Qwen3 Next 80B A3B Instruct',
+    id: 'qwen/qwen3-next-80b-a3b-instruct',
+    maxOutput: 65_536,
+    pricing: {
+      units: [
+        { name: 'textInput', rate: 0.15, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textOutput', rate: 1.5, strategy: 'fixed', unit: 'millionTokens' },
+      ],
+    },
+    type: 'chat',
+  },
+  {
+    contextWindowTokens: 4096,
+    displayName: 'Qwen MT Plus',
+    id: 'qwen/qwen-mt-plus',
+    maxOutput: 2048,
+    pricing: {
+      units: [
+        { name: 'textInput', rate: 0.25, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textOutput', rate: 0.75, strategy: 'fixed', unit: 'millionTokens' },
+      ],
+    },
+    type: 'chat',
+  },
   {
     abilities: {
       functionCall: true,
@@ -41,13 +87,13 @@ const novitaChatModels: AIChatModelCard[] = [
     abilities: {
       functionCall: true,
     },
-    contextWindowTokens: 262_144,
+    contextWindowTokens: 65_536,
     displayName: 'Qwen3 Coder 480B A35B Instruct',
     id: 'qwen/qwen3-coder-480b-a35b-instruct',
     pricing: {
       units: [
-        { name: 'textInput', rate: 0.35, strategy: 'fixed', unit: 'millionTokens' },
-        { name: 'textOutput', rate: 1.5, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textInput', rate: 0.29, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textOutput', rate: 1.2, strategy: 'fixed', unit: 'millionTokens' },
       ],
     },
     type: 'chat',
@@ -83,6 +129,24 @@ const novitaChatModels: AIChatModelCard[] = [
     },
     type: 'chat',
   },
+  {
+    abilities: {
+      functionCall: true,
+      reasoning: true,
+      vision: true,
+    },
+    contextWindowTokens: 65_536,
+    displayName: 'GLM-4.5V',
+    id: 'zai-org/glm-4.5v',
+    maxOutput: 16_384,
+    pricing: {
+      units: [
+        { name: 'textInput', rate: 0.6, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textOutput', rate: 1.8, strategy: 'fixed', unit: 'millionTokens' },
+      ],
+    },
+    type: 'chat',
+  },
   {
     abilities: {
       functionCall: true,
@@ -91,6 +155,7 @@ const novitaChatModels: AIChatModelCard[] = [
     contextWindowTokens: 131_072,
     displayName: 'GLM-4.5',
     id: 'zai-org/glm-4.5',
+    maxOutput: 98_304,
     pricing: {
       units: [
         { name: 'textInput', rate: 0.6, strategy: 'fixed', unit: 'millionTokens' },
@@ -103,9 +168,10 @@ const novitaChatModels: AIChatModelCard[] = [
     abilities: {
       functionCall: true,
     },
-    contextWindowTokens: 262_144,
+    contextWindowTokens: 131_072,
     displayName: 'Qwen3 235B A22B Instruct 2507',
     id: 'qwen/qwen3-235b-a22b-instruct-2507',
+    maxOutput: 16_384,
     pricing: {
       units: [
         { name: 'textInput', rate: 0.15, strategy: 'fixed', unit: 'millionTokens' },
@@ -717,6 +783,7 @@ const novitaChatModels: AIChatModelCard[] = [
     contextWindowTokens: 32_000,
     displayName: 'Qwen2.5 72B Instruct',
     id: 'qwen/qwen-2.5-72b-instruct',
+    maxOutput: 8192,
     pricing: {
       units: [
         { name: 'textInput', rate: 0.38, strategy: 'fixed', unit: 'millionTokens' },
@@ -770,7 +837,6 @@ const novitaChatModels: AIChatModelCard[] = [
     },
     contextWindowTokens: 32_768,
     displayName: 'Qwen2.5 VL 72B Instruct',
-    enabled: true,
     id: 'qwen/qwen2.5-vl-72b-instruct',
     pricing: {
       units: [

package/packages/model-bank/src/aiModels/qwen.ts

@@ -48,6 +48,92 @@ const qwenChatModels: AIChatModelCard[] = [
     },
     type: 'chat',
   },
+  {
+    abilities: {
+      reasoning: true,
+    },
+    contextWindowTokens: 131_072,
+    description: 'GLM-4.5系列模型是智谱AI专为智能体设计的混合推理模型，提供思考与非思考两种模式。',
+    displayName: 'GLM-4.5',
+    id: 'glm-4.5',
+    maxOutput: 16_384,
+    pricing: {
+      currency: 'CNY',
+      units: [
+        {
+          lookup: {
+            prices: {
+              '[0, 32_000]': 3,
+              '[32_000, infinity]': 4,
+            },
+            pricingParams: ['textInputRange'],
+          },
+          name: 'textInput',
+          strategy: 'lookup',
+          unit: 'millionTokens',
+        },
+        {
+          lookup: {
+            prices: {
+              '[0, 32_000]': 14,
+              '[32_000, infinity]': 16,
+            },
+            pricingParams: ['textInputRange'],
+          },
+          name: 'textOutput',
+          strategy: 'lookup',
+          unit: 'millionTokens',
+        },
+      ],
+    },
+    settings: {
+      extendParams: ['enableReasoning', 'reasoningBudgetToken'],
+    },
+    type: 'chat',
+  },
+  {
+    abilities: {
+      reasoning: true,
+    },
+    contextWindowTokens: 131_072,
+    description: 'GLM-4.5系列模型是智谱AI专为智能体设计的混合推理模型，提供思考与非思考两种模式。',
+    displayName: 'GLM-4.5-Air',
+    id: 'glm-4.5-air',
+    maxOutput: 16_384,
+    pricing: {
+      currency: 'CNY',
+      units: [
+        {
+          lookup: {
+            prices: {
+              '[0, 32_000]': 0.8,
+              '[32_000, infinity]': 1.2,
+            },
+            pricingParams: ['textInputRange'],
+          },
+          name: 'textInput',
+          strategy: 'lookup',
+          unit: 'millionTokens',
+        },
+        {
+          lookup: {
+            prices: {
+              '[0, 32_000]': 6,
+              '[32_000, infinity]': 8,
+            },
+            pricingParams: ['textInputRange'],
+          },
+          name: 'textOutput',
+          strategy: 'lookup',
+          unit: 'millionTokens',
+        },
+      ],
+    },
+    settings: {
+      extendParams: ['enableReasoning', 'reasoningBudgetToken'],
+    },
+    type: 'chat',
+  },
   {
     abilities: {
       functionCall: true,
@@ -583,7 +669,7 @@ const qwenChatModels: AIChatModelCard[] = [
       search: true,
     },
     config: {
-      deploymentName: 'qwen-plus-2025-07-28',
+      deploymentName: 'qwen-plus-2025-09-11',
     },
     contextWindowTokens: 1_000_000,
     description: '通义千问超大规模语言模型增强版，支持中文、英文等不同语言输入。',
@@ -627,7 +713,6 @@ const qwenChatModels: AIChatModelCard[] = [
               '[0, 128_000]_[false]': 2,
               '[0, 128_000]_[true]': 8,
               '[128_000, 256_000]_[false]': 20,
-
               '[128_000, 256_000]_[true]': 24,
               '[256_000, infinity]_[false]': 48,
               '[256_000, infinity]_[true]': 64,

package/packages/model-bank/src/aiModels/siliconcloud.ts

@@ -47,6 +47,25 @@ const siliconcloudChatModels: AIChatModelCard[] = [
     },
     type: 'chat',
   },
+  {
+    abilities: {
+      functionCall: true,
+    },
+    contextWindowTokens: 131_072,
+    description:
+      'Ling-mini-2.0 是一款基于 MoE 架构的小尺寸高性能大语言模型。它拥有 16B 总参数，但每个 token 仅激活 1.4B（non-embedding 789M），从而实现了极高的生成速度。得益于高效的 MoE 设计与大规模高质量训练数据，尽管激活参数仅为 1.4B，Ling-mini-2.0 依然在下游任务中展现出可媲美 10B 以下 dense LLM 及更大规模 MoE 模型的顶尖性能',
+    displayName: 'Ling Mini 2.0',
+    id: 'inclusionAI/Ling-mini-2.0',
+    pricing: {
+      currency: 'CNY',
+      units: [
+        { name: 'textInput', rate: 0.5, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textOutput', rate: 2, strategy: 'fixed', unit: 'millionTokens' },
+      ],
+    },
+    releasedAt: '2025-09-09',
+    type: 'chat',
+  },
   {
     abilities: {
       functionCall: true,
@@ -223,10 +242,51 @@ const siliconcloudChatModels: AIChatModelCard[] = [
     type: 'chat',
   },
   {
+    abilities: {
+      functionCall: true,
+    },
+    contextWindowTokens: 262_144,
+    description:
+      'Kimi K2-Instruct-0905 是 Kimi K2 最新、最强大的版本。它是一款顶尖的混合专家（MoE）语言模型，拥有 1 万亿的总参数和 320 亿的激活参数。该模型的主要特性包括：增强的智能体编码智能，在公开基准测试和真实世界的编码智能体任务中表现出显著的性能提升；改进的前端编码体验，在前端编程的美观性和实用性方面均有进步。',
+    displayName: 'Kimi K2 0905',
+    id: 'moonshotai/Kimi-K2-Instruct-0905',
+    pricing: {
+      currency: 'CNY',
+      units: [
+        { name: 'textInput', rate: 4, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textOutput', rate: 16, strategy: 'fixed', unit: 'millionTokens' },
+      ],
+    },
+    releasedAt: '2025-09-05',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+    },
+    contextWindowTokens: 262_144,
+    description:
+      'Kimi K2-Instruct-0905 是 Kimi K2 最新、最强大的版本。它是一款顶尖的混合专家（MoE）语言模型，拥有 1 万亿的总参数和 320 亿的激活参数。该模型的主要特性包括：增强的智能体编码智能，在公开基准测试和真实世界的编码智能体任务中表现出显著的性能提升；改进的前端编码体验，在前端编程的美观性和实用性方面均有进步。',
+    displayName: 'Kimi K2 0905 (Pro)',
+    id: 'Pro/moonshotai/Kimi-K2-Instruct-0905',
+    pricing: {
+      currency: 'CNY',
+      units: [
+        { name: 'textInput', rate: 4, strategy: 'fixed', unit: 'millionTokens' },
+        { name: 'textOutput', rate: 16, strategy: 'fixed', unit: 'millionTokens' },
+      ],
+    },
+    releasedAt: '2025-09-05',
+    type: 'chat',
+  },
+  {
+    abilities: {
+      functionCall: true,
+    },
     contextWindowTokens: 131_072,
     description:
       'Kimi K2 是一款具备超强代码和 Agent 能力的 MoE 架构基础模型，总参数 1T，激活参数 32B。在通用知识推理、编程、数学、Agent 等主要类别的基准性能测试中，K2 模型的性能超过其他主流开源模型。',
-    displayName: 'Kimi K2 Instruct',
+    displayName: 'Kimi K2 0711',
     id: 'moonshotai/Kimi-K2-Instruct',
     pricing: {
       currency: 'CNY',
@@ -239,10 +299,13 @@ const siliconcloudChatModels: AIChatModelCard[] = [
     type: 'chat',
   },
   {
+    abilities: {
+      functionCall: true,
+    },
     contextWindowTokens: 131_072,
     description:
       'Kimi K2 是一款具备超强代码和 Agent 能力的 MoE 架构基础模型，总参数 1T，激活参数 32B。在通用知识推理、编程、数学、Agent 等主要类别的基准性能测试中，K2 模型的性能超过其他主流开源模型。',
-    displayName: 'Kimi K2 Instruct (Pro)',
+    displayName: 'Kimi K2 0711 (Pro)',
     id: 'Pro/moonshotai/Kimi-K2-Instruct',
     pricing: {
       currency: 'CNY',

package/packages/model-bank/src/aiModels/volcengine.ts

@@ -35,17 +35,18 @@ const doubaoChatModels: AIChatModelCard[] = [
       functionCall: true,
     },
     config: {
-      deploymentName: 'kimi-k2-250711',
+      deploymentName: 'kimi-k2-250905',
     },
-    contextWindowTokens: 256_000,
+    contextWindowTokens: 131_072,
     description:
       'Kimi-K2 是一款Moonshot AI推出的具备超强代码和 Agent 能力的 MoE 架构基础模型，总参数 1T，激活参数 32B。在通用知识推理、编程、数学、Agent 等主要类别的基准性能测试中，K2 模型的性能超过其他主流开源模型。',
     displayName: 'Kimi K2',
     id: 'kimi-k2',
-    maxOutput: 16_384,
+    maxOutput: 32_768,
     pricing: {
       currency: 'CNY',
       units: [
+        { name: 'textInput_cacheRead', rate: 0.8, strategy: 'fixed', unit: 'millionTokens' },
         { name: 'textInput', rate: 4, strategy: 'fixed', unit: 'millionTokens' },
         { name: 'textOutput', rate: 16, strategy: 'fixed', unit: 'millionTokens' },
       ],

package/packages/model-runtime/src/utils/modelParse.ts

@@ -53,7 +53,7 @@ export const MODEL_LIST_CONFIGS = {
       'qwen2.5',
       'qwen3',
     ],
-    reasoningKeywords: ['qvq', 'qwq', 'qwen3', '!-instruct-', '!-coder-'],
+    reasoningKeywords: ['qvq', 'qwq', 'qwen3', '!-instruct-', '!-coder-', '!-max-'],
     visionKeywords: ['qvq', 'vl'],
   },
   v0: {
@@ -303,9 +303,9 @@ const processModelCard = (
     )
       ? 'image'
       : isKeywordListMatch(
-            model.id.toLowerCase(),
-            EMBEDDING_MODEL_KEYWORDS.map((k) => k.toLowerCase()),
-          )
+        model.id.toLowerCase(),
+        EMBEDDING_MODEL_KEYWORDS.map((k) => k.toLowerCase()),
+      )
         ? 'embedding'
         : 'chat');