这是 时隔21个月 以来的最大版本的更新。 金磊 发自 凹非寺 量子位 | 公众号 QbitAI 在 Seedance 2.0 和 Seedream 5.0 Lite ，一波接一波爆火之后，豆包把完全体拿出来了—— 豆包大模型2.0 。 像Seedance 2.0已经成为全民玩转的AI，我们也试着做了一个视频： 短短5秒钟，效果确实是足够逼真。 也难怪老外也开始研究怎么注册中国手机号来体验了…… 再如 Seedream 5.0 Lite ，首次支持联网检索，生成的图片也达到了商业化的水平： 而就在今天，在视觉模型火爆之后，豆包终于把那个最核心的大脑拿出来了—— 豆包大模型2.0 。 整体来看，这次豆包大模型2.0在多模态理解、企业级Agent、推理和代码能力上都有了不少的提升： 更直观的提升，体现在榜单测评中。 例如在MathVista、MathVision、MathKangaroo、MathCanvas等数学推理基准上达到业界最优水平。同时，在 LogicVista、VisuLogic 等视觉解谜与逻辑推理基准上，Seed2.0 Pro得分较Seed1.8显著提升。 更强多模态理解：在多模态感知、高精度文字 ...

中新社记者 王梦瑶 春节渐近，中国各大高铁车站里的归家人越来越多，随身携带的行李却越来越少。 在北京南站的候车厅里，旅客们不再大包小裹出行，许多人轻装简从，一个行李箱加一个书包成了标 配，有的男士只背着双肩包出行。 今年和孩子从天津西站回青岛过年的肖晓旭告诉中新社记者，他们的行李都已经提前通过快递寄回了 家。肖晓旭说，她已经习惯了轻装出行，尤其在有了孩子以后，更加倾向减少行李，这样在路上就可以 只顾小孩，自己更轻松省心。"这会儿我人还在车上，但行李已经到家了。"她说。 2月13日，肖晓旭和儿子在车站，母子二人仅携带少量随身背包。 中新社记者 王梦瑶 摄 中新社北京2月14日电 题：物"流"快递，中国人春节旅途轻松了 这种出行习惯的形成与中国快递网络的日益完善密不可分。"十四五"以来，中国建成了世界上规模最 大、受益人数最多的寄递网络。农村和边境地区邮政通达广度深度明显提升，营业网点数量较5年前增 长近1.5倍，全国边境自然村全部实现通邮。快递物流的触角向更远更深处延伸，更多人的需求得以释 放。 在过年带回家的行李中，给亲人朋友带的年货礼品占了不少比例。杨岳说，不少人会在节前购买新疆、 内蒙古、甘肃的羊肉尝鲜， ...

Core Insights - Doubao model has officially entered its 2.0 phase, focusing on systematic optimizations for large-scale production environments [1][3] Group 1: Doubao 2.0 Features - Doubao 2.0 includes Pro, Lite, Mini, and Code models, designed to adapt to various business scenarios [3] - Doubao 2.0 Pro targets deep reasoning and long-chain task execution, competing directly with GPT 5.2 and Gemini 3 Pro [3] - Doubao 2.0 Lite balances performance and cost, surpassing the previous generation Doubao 1.8 in overall capabilities [3] - Doubao 2.0 Mini is aimed at low-latency, high-concurrency, and cost-sensitive scenarios [3] - The Code version is specifically designed for programming tasks and works best in conjunction with TRAE [3] Group 2: Product Launch and User Experience - Doubao 2.0 Pro is now available on Doubao App, desktop, and web versions, allowing users to experience the "expert" mode [3] - The Code version has been integrated with AI programming product TRAE, and API services for Doubao 2.0 models are now available for enterprises and developers via Volcano Engine [3] - Seedance 2.0, a video generation model, has been officially released and is fully integrated into Doubao and Jimeng products, with a user experience center launched for testing [3][4] Group 3: Seedance 2.0 Capabilities - Seedance 2.0 allows users to generate 5 or 10-second videos by entering prompts in the Doubao App [4] - The model supports synchronized audio-visual generation, multi-shot long narratives, and controllable multi-modal outputs [4] - The company acknowledges that Seedance 2.0 is not yet perfect and aims to improve the alignment between large models and human feedback for better audio-visual production tools [4]

Core Insights - The article discusses the intense competition among internet giants in the AI large model sector, highlighting the ambitions of companies to establish their AI applications as the primary traffic entry point [4][23] - DeepSeek, founded by Liang Wenfeng, emerged as a significant player in the AI landscape with its R1 model, which was launched at a surprisingly low cost, challenging the perception of high investment requirements for top-tier models [14][31] - Despite the competitive environment, DeepSeek has maintained a low profile, with recent updates suggesting a potential new model release, V4, but with no official confirmation [26][27] Industry Competition - Major companies are aggressively distributing cash incentives to attract users, with Tencent offering 1 billion yuan, Baidu 500 million yuan, and Alibaba 3 billion yuan, indicating a fierce battle for user engagement [25] - The launch of new models by ByteDance and Alibaba, including the 2.0 versions of their respective models, reflects a rapid evolution in AI capabilities and competition [8][25] - The article notes a peculiar competitive dynamic where companies are responding to each other's moves, creating a sense of mutual awareness in the market [8][25] DeepSeek's Position - DeepSeek's recent updates include an increase in context window length from 128K tokens to 1 million tokens, suggesting advancements in their technology [26] - The company continues to recruit talent despite a slowdown in hiring across the industry, indicating its commitment to innovation and development [27] - Liang Wenfeng's vision for DeepSeek is to lead in AI research and development, aiming to create a general-purpose AI that goes beyond existing models [31][32] User Engagement and Market Dynamics - The article emphasizes the importance of addressing user needs in the AI sector, with companies like DeepSeek beginning to focus on consumer-facing products [33] - The competition is framed as a quest to meet real user demands, which will determine the leading players in the AI landscape [36] - The article concludes that the current battle among internet giants is crucial for defining the next decade of internet order, highlighting the strategic significance of user engagement in AI applications [36]

Core Insights - ByteDance has launched the Doubao-Seed-2.0 series, optimizing for large-scale production environments to handle complex real-world tasks [1] - The Doubao-2.0 Pro flagship version has achieved top-tier performance in various competitions, surpassing Gemini 3 Pro in mathematical reasoning and problem-solving capabilities [1] - Doubao-2.0 enhances knowledge coverage in niche areas and performs well in scientific knowledge tests, ranking alongside Gemini 3 Pro and GPT 5.2 [1] Group 1 - Doubao-2.0 has upgraded its multimodal understanding capabilities, achieving industry-best performance in visual reasoning, spatial perception, and long-context understanding [1] - The model has improved its understanding of time series and motion perception, enabling real-time video analysis and environmental interaction for applications in fitness guidance and companionship [2] - The Doubao-2.0 Pro has been launched on various platforms, including the Doubao App and web client, allowing users to experience its capabilities in "expert" mode [2] Group 2 - The pricing for Doubao-2.0 Pro is set based on input length, with costs of 3.2 yuan per million tokens for inputs under 32k and 16 yuan per million tokens for outputs, offering a significant cost advantage over competitors [2] - Doubao-2.0 Lite provides excellent cost-performance, with input pricing at only 0.6 yuan per million tokens, surpassing the previous generation model Doubao-1.8 [2]

Core Insights - Doubao 2.0 has officially entered its second phase, focusing on systematic optimizations for large-scale production environments, enhancing capabilities in efficient inference, multimodal understanding, and complex instruction execution to better tackle real-world complex tasks [1] Group 1: Product Features - The Doubao 2.0 series includes three general-purpose agent models (Pro, Lite, Mini) and a Code model, designed to flexibly adapt to various business scenarios [2] - Doubao 2.0 Pro targets deep reasoning and long-chain task execution, directly competing with GPT 5.2 and Gemini 3 Pro [2] - Doubao 2.0 Lite balances performance and cost, with comprehensive capabilities surpassing the previous generation model Doubao 1.8 [2] - Doubao 2.0 Mini is aimed at low-latency, high-concurrency, and cost-sensitive scenarios [2] - The Code version (Doubao-Seed-2.0-Code) is specifically designed for programming scenarios and performs better when used in conjunction with TRAE [2]

Core Insights - Doubao model has officially entered its 2.0 phase, focusing on systematic optimizations for large-scale production environments [1][3] Group 1: Doubao 2.0 Features - Doubao 2.0 includes Pro, Lite, Mini, and Code models, designed to adapt to various business scenarios [3] - Doubao 2.0 Pro targets deep reasoning and long-chain task execution, competing directly with GPT 5.2 and Gemini 3 Pro [3] - Doubao 2.0 Lite balances performance and cost, surpassing the previous generation model Doubao 1.8 in overall capabilities [3] - Doubao 2.0 Mini is tailored for low latency, high concurrency, and cost-sensitive scenarios [3] - The Code version is specifically designed for programming tasks and works best in conjunction with TRAE [3] Group 2: Product Launch and User Experience - Doubao 2.0 Pro is now available on Doubao App, desktop, and web versions, allowing users to experience the "expert" mode for conversations [3] - The Code version has been integrated with AI programming product TRAE, and API services for Doubao 2.0 models are now available for enterprises and developers [3] - Seedance 2.0, a video generation model, has been launched and is fully integrated into Doubao and Jimeng products, with a user experience center now available [3][4] Group 3: Seedance 2.0 Capabilities - Seedance 2.0 allows users to generate 5 or 10-second videos by inputting relevant prompts [4] - The model supports original sound and image synchronization, multi-shot long narratives, and controllable multi-modal generation [4] - The company acknowledges that Seedance 2.0 is not yet perfect and aims to improve the alignment between large models and human feedback for better audio-visual production tools [4]

21世纪经济报道记者闫硕 这个春节，元宝、千问、豆包掀起AI红包大战，主打C端健康场景的蚂蚁阿福也迅速入局。另一方面， 短视频平台的爆款内容让部分AI产品成功出圈，大模型这股风正加速吹向下沉市场。 在皖北一座小县城，用大模型当"健康顾问"已并不鲜见。49岁的秀梅前段时间体检时查出甲状腺结节， 跑了好几家医院，医生都建议静养观察，但她心里始终不踏实。侄子给她推荐了一款垂类大模型，自 此，她便频繁咨询，看着"不必过于担心"的回复，这款产品成了她康复路上的一颗"定心丸"。 后来，侄子推荐给她一款大模型APP，"我最开始问的是'良性甲状腺结节会不会转成恶性的'，输入问题 后几秒便有了回复。"秀梅说，看着"恶变概率非常低，通常在5%以下，很多研究甚至认为低于1%"的回 复，而且其中还附了多份权威资料链接，她松了一口气，反复读了好多遍。此后她又追问康复期间的各 类注意事项，AI的答复总能接住她的细碎担忧。 用户对大模型的应用，早已不局限于健康咨询场景。记者通过走访线下药店了解到，有不少人已经开始 拿着大模型推荐的药品进行咨询、购买。"不少人过年期间存有忌讳就医的习俗，即便出现小病小痛也 不愿前往医院，大多会选择自行购药缓 ...

Core Insights - ByteDance launched Doubao-Seed-2.0 series on February 14, enhancing its large model capabilities for complex real-world tasks [1] - Doubao-2.0 Pro flagship version achieved top scores in IMO, CMO math competitions, and ICPC programming contests, surpassing Gemini 3 Pro in Putnam benchmark tests [1] - The model has improved knowledge coverage in niche areas and performed well in scientific knowledge tests, ranking comparably with Gemini 3 Pro and GPT 5.2 [1] - Doubao-2.0 has upgraded its multimodal understanding capabilities, excelling in visual reasoning, spatial perception, and long-context understanding tests [1] Application and Features - Doubao-2.0 enhances understanding of time series and motion perception, enabling real-time video analysis and environmental interaction for applications in fitness guidance, fashion advice, and companionship [2] - The model's agent capabilities achieved top scores in instruction following, tool invocation, and Search Agent evaluations, with a record score of 54.2 in HLE-Text [4] - Doubao-2.0 Pro is available on Doubao App, PC client, and web version, with a cost-effective pricing model based on input length [4] - The pricing for Doubao-2.0 Pro is set at 3.2 yuan per million tokens for inputs under 32k and 16 yuan per million tokens for outputs, offering a significant cost advantage over Gemini 3 Pro and GPT 5.2 [4] - Doubao-2.0 Lite offers high cost-performance, with input pricing at only 0.6 yuan per million tokens, surpassing the previous generation model Doubao-1.8 [4]

Core Insights - ByteDance announced the release of Doubao Model 2.0, which includes Pro, Lite, Mini, and Code models, optimized for large-scale production environments to enhance multimodal understanding, complex instruction execution, and long-range task processing capabilities [1] Group 1: Model Features - Doubao 2.0 Pro targets deep reasoning and long-chain tasks, competing with GPT 5.2 and Gemini 3 Pro [1] - Doubao 2.0 Lite surpasses the previous main model Doubao 1.8 in overall capabilities [1] - Doubao 2.0 Mini is designed for low latency, high concurrency, and cost-sensitive scenarios [1] - The Code model is specifically developed for programming scenarios and can integrate with AI programming product TRAE [1] Group 2: Performance Metrics - Doubao 2.0 has achieved state-of-the-art (SOTA) levels in multimodal understanding, excelling in visual reasoning, perception, spatial reasoning, and long-context understanding tasks [1] - In dynamic scene evaluations like TVBench and EgoTempo, Doubao 2.0 outperformed human benchmarks [1] - The model demonstrated excellent performance in long video understanding across multiple streaming real-time question-answering benchmarks [1] Group 3: Competitive Edge - Doubao 2.0 Pro scored higher than GPT 5.2 in scientific evaluations such as SuperGPQA and HealthBench [1] - The model achieved gold medal results in IMO, CMO mathematics competitions, and ICPC programming contests [1] - The reasoning cost of Doubao 2.0 Pro is approximately an order of magnitude lower than that of leading industry models [1] Group 4: Availability - Doubao 2.0 Pro is now available on Doubao App, desktop, and web versions, with users able to experience it in "Expert" mode [2] - The Code model has been integrated into the Chinese version of TRAE [2] - Volcano Engine has launched API services for the Doubao 2.0 series models for enterprise and developer use [2]