千问发布多模态新模型 追平Gemini Flash定价不足两成

On September 19, 2026, Qwen officially launched Qwen3.8-Omni-Flash, its first multimodal model purpose-built for AI agents. The model matches the performance of Google’s comparable offering at less than 20% of the price, and is now open to public access, allowing regular users to complete a wide range of audio and video processing tasks at low cost with the model and its supporting tools.

1. Core capabilities cover high-frequency daily use cases. The model can process audio and video content simultaneously, and autonomously call tools to complete tasks such as vlog editing, short video translation, and film and television content summarization. It features a 1 million-token context window, with audio and video processing performance on par with Google Gemini 3.8 Flash, meeting the needs of most personal audio and video processing scenarios.

2. It offers a substantial cost advantage. The model API is priced at $0.15 per million input tokens and $0.47 per million output tokens. According to official estimates, audio input costs less than $0.01 per hour, while processing a 720p video with audio, sampled at one frame per second, costs roughly $0.20—far lower than Google’s comparable product, which is set to double its price on January 1, 2027.

3. It has low adoption barriers. The model is now accessible via three channels: Qwen Studio, Qwen Cloud, and the official API. Qwen has also released the open-source Qwen-MM-Plugins component, which adds practical capabilities such as video editing and speaker recognition to mainstream agent products. The supporting Qwen-Live Harness tool further supports real-time interaction via camera and microphone, allowing users to choose the access method that best fits their needs.

The high-value multimodal model launched by Qwen provides brands with a new tool option and strategic reference for advancing AI marketing, optimizing products and services, and managing costs. Brands can explore its application value based on their own needs.

1. Its pricing strategy offers useful reference. The model’s audio and video processing performance matches Google Gemini 3.8 Flash, yet its API is priced at less than 20% of Google’s current offering, which will double in price in January 2027. This high-value pricing logic can inform brands’ own AI service pricing and supply chain cost management.

2. It creates new support for content production and product iteration. The model can autonomously complete audio and video processing tasks such as vlog editing, short video translation, and content summarization, while its supporting open-source plugins enable capabilities including PDF-based video notes and reusable workflows. Brands can leverage these tools to cost-effectively produce marketing short videos at scale, adapt content for multiple languages, and improve user-facing audio and video AI services.

3. It provides flexible access channels. The model is now available through Qwen Studio, Qwen Cloud, and the official API, allowing brands to select the integration method aligned with their technical capabilities and business needs, and rapidly deploy relevant applications.

The cost-effective multimodal model launched by Qwen provides low-threshold tool support for sellers of all types to reduce costs, improve efficiency, and identify new business growth opportunities. Sellers should pay close attention to its potential use cases.

1. It can significantly lower content production costs. The model’s audio and video processing costs are far below those of comparable overseas products: audio input costs less than $0.01 per hour, while processing a 720p video with audio at one frame per second costs around $0.20. In addition, Google’s comparable product will raise prices in 2027. By adopting this model for short video editing, content translation, livestream content summarization and other tasks, sellers can sharply cut content production spending.

2. It opens up clear new business scenario opportunities. The model can autonomously handle tasks such as vlog editing, short video translation, and audio and video content summarization. Its supporting open-source components enable capabilities including speaker recognition and reusable workflows, while companion tools support real-time camera and microphone interaction. Sellers can use these features to mass-produce multilingual product-promotion short videos, quickly review livestream content, and build real-time audio and video interactive customer services to unlock new revenue streams.

3. It has low barriers to deployment. The model is now open to access through multiple official channels, and its supporting open-source components can be directly integrated with mainstream agent products, allowing sellers to test and launch relevant applications quickly without heavy technical investment.

The cost-effective multimodal large model and supporting tools released by Qwen provide practical, low-cost technical support for manufacturing factories advancing digital transformation and expanding into e-commerce. Factories can explore related business opportunities based on the publicly available capabilities.

1. Its multimodal processing capabilities can cover a range of operational needs. The model features a 1 million-token context window, can process audio and video simultaneously, and can autonomously complete tasks such as vlog editing, short video translation, and content summarization. Official open-source components further provide speaker recognition, PDF-based video notes, and reusable workflow capabilities, while supporting tools enable real-time interaction via cameras and microphones. Factories can build applications tailored to their own production and operational scenarios on the basis of these public capabilities, reducing labor costs for related work.

2. It can substantially cut content production costs for online operations. The model’s audio and video processing costs are extremely low, far below Google’s comparable product, which will double in price in 2027. When factories develop e-commerce channels, produce product promotional short videos, or convert content into multiple languages for overseas markets, they can directly integrate this model to reduce content production expenses.

3. It has low adoption barriers. The model is now open through multiple channels including Qwen Studio, Qwen Cloud, and the official API, and its supporting open-source components can connect with mainstream agent products. Factories can quickly access and use the model without heavy R&D investment, easing the cost pressure of digital transformation.

The multimodal model and supporting open-source tools launched by Qwen point to a new direction for the practical deployment of multimodal agents for AI application service providers and enterprise service vendors, while also providing directly reusable technical components that can address relevant customer pain points.

1. The industry technology trend is clear. Multimodal models built for AI agents represent a core track for current AI deployment. The model released by Qwen can process audio and video simultaneously, autonomously call tools, and support a 1 million-token context window, with performance matching Google Gemini 3.8 Flash at an extremely low price point. This reflects the trend toward high-value, inclusive deployment of multimodal large models.

2. It aligns closely with customer pain points. At present, most enterprise customers face challenges including high audio and video content processing costs, high barriers to multimodal agent development, and insufficient real-time interaction capabilities. The model’s audio and video processing costs are far below those of comparable overseas products; its supporting open-source Qwen-MM-Plugins component adds capabilities such as video editing, speaker recognition, and reusable workflows; and its companion tools support real-time audio and video calls. Altogether, it can serve as a foundational technical component for service providers building customized solutions for clients.

3. Integration and cooperation are convenient. Qwen opens its capabilities through multiple channels, including Qwen Studio, Qwen Cloud, and the official API, allowing service providers to choose flexible integration methods based on their solution needs and quickly launch related services.

Qwen’s launch of a high-value multimodal model, its release of supporting open-source tools, and its opening of multiple access channels provide a reference for various internet platforms and AI service platforms to improve their services, reduce operating costs, and mitigate related risks.

1. It points to a clear direction for platform capability upgrades. At present, merchants and users on platforms have strong demand for audio and video processing, intelligent tools, and real-time interaction. The model supports high-frequency audio and video tasks such as vlog editing, short video translation, and content summarization, while its supporting components can connect with mainstream agents and support real-time audio and video calls. By integrating this capability, platforms can provide cost-effective supporting tools for their users to meet those demands.

2. It enables controllable costs and risks. The leading overseas comparable product is currently priced at more than five times Qwen’s model and will double its price in January 2027. If platforms have previously relied on overseas models to provide related services, switching to this domestic model can significantly reduce operating costs while mitigating risks from overseas service price hikes and supply chain instability.

3. Its ecosystem operation model offers useful lessons. Qwen’s approach of opening capabilities through multiple channels and releasing open-source components to lower access barriers can provide reference for platforms recruiting developers and building AI application ecosystems. Platforms can design corresponding operating rules around the model’s capabilities to enrich their on-platform service ecosystems.

Qwen’s release of a high-value multimodal model matching the performance of leading overseas products marks an important new development in the domestic deployment of AI large models, and provides a fresh case for research on the commercialization of the multimodal agent track and the substitution of domestic large models.

1. New characteristics have emerged in industrial technology and competition. The newly launched Qwen3.8-Omni-Flash is a multimodal model purpose-built for AI agents. It can process audio and video simultaneously, autonomously call tools, and features a 1 million-token context window, with performance matching Google Gemini 3.8 Flash, yet its API is priced at less than 20% of the latter’s current price. Google’s comparable product will also double in price in early 2027. This demonstrates that domestic large models have built strong price competitiveness while matching the performance of leading overseas products, substantially lowering the cost threshold for multimodal large model deployment.

2. It provides a new case for large model commercial ecosystem building. In addition to opening model APIs, Qwen has simultaneously launched the open-source Qwen-MM-Plugins component to connect with multiple mainstream agent products, paired with supporting tools for real-time interaction. By combining base model access with open-source components to complete the ecosystem, the company has lowered developer access barriers, offering a new reference for research on large model commercialization paths.

3. It offers new insights for industrial development. The mature deployment of cost-effective domestic multimodal models provides a feasible path for China’s AI industry to reduce dependence on overseas products and control application deployment costs. Relevant industrial support policies can be optimized in a targeted manner around the deployment and ecosystem building of such high-value foundation models.