Core Insights - The article discusses the advancements in AI technology, particularly focusing on the launch of Doubao Model 1.8 and Seedance 1.5 pro by Huoshan Engine, highlighting their capabilities in multi-modal understanding and content creation [3][4][6]. Group 1: Doubao Model 1.8 - Doubao Model 1.8 has significantly enhanced its multi-modal understanding capabilities, increasing video frame understanding from 640 to 1280 frames, which supports various applications like online education and industrial quality inspection [4][5]. - The model's tool usage and complex instruction adherence capabilities have been improved, making it suitable for enterprise-level tasks that require planning and execution [5][6]. - Doubao Model 1.8 supports a context window of 256K and offers API management for context, optimizing performance while reducing costs [5][6]. Group 2: Seedance 1.5 pro - Seedance 1.5 pro introduces a native audio-video joint generation architecture, allowing for real-time synchronization of audio and visual elements, enhancing the realism of generated videos [6][7]. - The model supports multi-language dialogue and precise lip-syncing, significantly improving the global creative potential of video content [7][8]. - A "Draft Sample" feature will be launched to allow creators to preview low-resolution samples, increasing efficiency by 65% and reducing ineffective production costs by 60% [8]. Group 3: AI Cloud-Native Architecture - Huoshan Engine is transitioning to an AI cloud-native architecture to support the scaling of enterprise Agent applications, addressing challenges in identity management and system integration [9][10]. - The AgentKit platform has been upgraded to cover the entire lifecycle of Agent development, deployment, and management [9]. - The average number of intelligent agents per enterprise is expected to increase from dozens in 2024 to over 200 in 2025, with applications expanding from consumer entertainment to serious production scenarios [10].

Core Insights - The article highlights the announcements made at the ByteDance FORCE conference, including the launch of the Doubao model 1.8 and the Seedance 1.5 pro audio-video creation model, along with updates on Agent development tools [1][3]. Doubao Model 1.8 - The Doubao model 1.8 has achieved a daily token usage of over 50 trillion, marking a growth of over 10 times compared to the same period last year [3]. - The model has been optimized for multi-modal Agent scenarios, enhancing its tool invocation capabilities, complex instruction adherence, and OS Agent capabilities, improving task planning and execution [3][4]. - In visual understanding, the model's video frame comprehension has increased from 640 to 1280 frames, allowing for low-frame-rate understanding of long videos and high-frame-rate comprehension of key segments [4]. Seedance 1.5 Pro - The Seedance 1.5 pro model is designed for video creation, featuring cinematic narrative tension and advanced motion detail capture, along with breakthroughs in audio-visual synchronization technology [5][6]. - It employs a native audio-video joint generation architecture, achieving millisecond-level audio-visual sync and supporting multiple languages and dialects [6]. - The upcoming "Draft Sample" feature will allow creators to generate low-resolution previews, enhancing overall efficiency by 65% and reducing ineffective creation costs by 60% [6]. AI Agent Development - The year has been termed the "AI Agent" year, with a shift towards model-centric AI cloud-native architectures to meet the demands of the Agent era [8]. - The AgentKit platform has been upgraded to cover the entire lifecycle of Agent development, deployment, and management, addressing core challenges faced by enterprises [8]. - The introduction of the HiAgent workstation aims to facilitate scalable management and application of Agents within enterprises [8]. - The "AI Savings Plan" offers tiered discounts for on-demand large model products, potentially saving enterprises up to 47% [8].

Core Insights - The era of artificial intelligence (AI) driven by intelligent agents is accelerating, leading to the evolution of cloud architecture from traditional IaaS, PaaS, and SaaS models to a model-centric AI cloud-native architecture [1] - In this new architecture, models become the core of software, and Model as a Service (MaaS) is identified as the optimal way to utilize these models [1] Group 1: AI Cloud Architecture - The traditional IT architecture is becoming ineffective as the AI era progresses, necessitating a shift to a model-centric cloud-native architecture [1] - The new architecture focuses on intelligent agents, with cloud platforms and middleware assembling tokens into intelligent agents for seamless communication [1] Group 2: Product Launch and Performance - Volcano Engine launched two new models: Doubao Model 1.8 and Doubao Video Generation Model 1.5 Pro, optimized for multi-modal intelligent agent scenarios [2] - The Doubao model family is now ranked among the top globally in multi-modal understanding and generation capabilities, with a daily token usage exceeding 50 trillion, making it the leading model in China and third globally [2] Group 3: Cost Reduction Initiatives - To reduce model usage costs for enterprise users, Volcano Engine introduced the "AI Savings Plan," which includes tiered discounts for all pay-as-you-go model products [2] - Users can save up to 47% on usage costs as their consumption increases under this new pricing model [2]

Core Insights - Volcano Engine officially launched Doubao Model 1.8 and Seedance 1.5 pro at the FORCE conference, with Doubao Model ranking among the top globally in multimodal understanding, generation capabilities, and agent abilities [1][3] - As of December, Doubao Model's daily token usage exceeded 50 trillion, marking a growth of over 10 times compared to the same period last year, with over 100 enterprise clients using more than 1 trillion tokens [1] Doubao Model 1.8 - Doubao Model 1.8 is optimized for multimodal agent scenarios, enhancing tool invocation, complex instruction adherence, and OS agent capabilities, improving task planning and execution [3] - The model's video understanding capability increased from 640 frames to 1280 frames per instance, allowing for low-frame-rate understanding of long videos and high-frame-rate analysis of key segments, applicable in online education and product quality inspection [3] Performance Metrics - Doubao Model 1.8 demonstrated competitive performance in various public evaluations, achieving top or near-top scores in visual reasoning, general visual question answering, spatial understanding, and video understanding tasks [5] - It leads globally in the BrowserComp general intelligence assessment and is approaching the top tier of foundational capabilities in mathematics and reasoning [5] Seedance 1.5 pro - Seedance 1.5 pro addresses the growing demand for video creation, featuring cinematic narrative tension and precise motion detail capture, with breakthroughs in audio-visual synchronization technology [6] - The model supports multi-language dialogue with accurate lip-syncing, covering various Chinese dialects, English, and minority languages, enhancing the realism and global creative potential of video content [6] - A new "Draft Sample" feature will allow creators to generate low-resolution previews, improving overall efficiency by 65% and reducing ineffective creation costs by 60% [6] AI Cloud-Native Architecture - Volcano Engine is upgrading its enterprise-level AI agent platform, AgentKit, to address challenges in agent deployment, identity management, model determinism, and system integration [8] - The HiAgent workstation will facilitate the large-scale management and application of agents by providing a unified AI task scheduling center and customizable intelligent agents [8] - An "AI Savings Plan" has been introduced, offering tiered discounts for on-demand large model products, potentially saving enterprises up to 47% [8]

Core Insights - Volcano Engine officially launched Doubao Model 1.8 and Seedance 1.5 pro at the FORCE Original Power Conference, with Doubao Model ranking among the top global models in multimodal understanding, generation capabilities, and agent capabilities [1][5] Doubao Model 1.8 - Doubao Model 1.8 has achieved a daily token usage exceeding 50 trillion, a growth of over 10 times compared to the same period last year [1][5] - More than 100 enterprise clients have accumulated over 1 trillion tokens in usage [1][5] - The model is optimized for multimodal agent scenarios, enhancing tool invocation, complex instruction adherence, and OS agent capabilities, improving task planning and execution [1][5] - In visual understanding, the model's video frame understanding capability has doubled from 640 frames to 1280 frames, supporting low-frame-rate understanding of long videos and high-frame-rate understanding of key segments [1][5] Seedance 1.5 pro - Seedance 1.5 pro addresses the growing demand for video creation, featuring cinematic narrative tension and precise motion detail capture, along with breakthroughs in audio-visual synchronization technology [2][6] - The model employs an innovative native audio-video joint generation architecture, achieving millisecond-level audio-visual synchronization and supporting multi-language dialogue with accurate lip-syncing [2][6] - A new "Draft Sample" feature will allow creators to generate low-resolution previews that closely match the final output, improving overall efficiency by 65% and reducing ineffective creation costs by 60% [2][6] - Individual users can experience the model on platforms like Doubao and Jimeng AI, while enterprise users can access it via Volcano Engine API starting December 23 [2][6] Agent Development - Volcano Engine has upgraded its enterprise-level AI Agent platform, AgentKit, covering the entire lifecycle of agent development, deployment, and management [3][7] - The platform aims to address core challenges in agent implementation, including identity and permission management, model determinism, and system integration [3][7] - The company introduced the first "AI Savings Plan" in the industry, offering tiered discounts for all pay-as-you-go large model products, enabling enterprises to save up to 47% on costs [3][7]