DeepSeek Open-Sources Multimodal Model; ChatGPT Ad Revenue Surpasses $1 Billion Annually

DeepSeek Releases Its First Multimodal Vision Model: A Major Addition to the Open-Source Ecosystem

In the late hours of August 31, DeepSeek officially open-sourced its first multimodal model, DeepSeek-V4-Flash-Vision-Exp, on Hugging Face under the permissive MIT License. The release includes model weights, a tokenizer, prompt encoding, and a minimal PyTorch inference reference implementation.

This marks a significant step for DeepSeek into the multimodal space. Compared to its previous models known primarily for language capabilities, the release of V4 Flash Vision is notable not only for unlocking visual understanding, but also for its "minimal inference implementation" design philosophy — developers can quickly validate and deploy using the reference code without setting up a complex engineering stack. The choice of MIT License further lowers the barrier to commercial adoption, and is expected to attract a large number of third-party applications and fine-tuning projects.

Strategically, DeepSeek's decision to enter the multimodal arena through open source represents direct competition against closed multimodal systems like OpenAI GPT-4o and Google Gemini, while also reinforcing its influence within the open-source community.

ChatGPT Ad Revenue Breaks $1 Billion: A Diversified Business Model Takes Shape

OpenAI announced this week that its advertising business has reached an annualized revenue run rate of $1 billion (approximately ¥6.736 billion RMB). This milestone was achieved roughly 200 days after launch — an impressively fast trajectory.

ChatGPT ads are currently shown to OpenAI Go subscribers and free-tier users, and have rolled out across more than 40 countries and regions. OpenAI positions advertising as a "fourth growth engine" alongside enterprise services, consumer subscriptions, and the API — marking a meaningful shift from single-source API dependency toward a diversified revenue structure.

It's worth noting that combining advertising with AI products is inherently tension-filled — the balance between user experience and monetization will continue to test OpenAI's product judgment. From a financial standpoint, however, $1 billion in annualized revenue for an ad business less than a year old confirms that real market demand exists.

AWS GovCloud Expands: Federal Agencies Can Now Access Anthropic, OpenAI, and Other Leading LLMs

Amazon Web Services announced an expansion of GovCloud AI model availability, allowing U.S. federal agencies to access leading models from Anthropic, OpenAI, Meta, and Amazon's own Nova series through Amazon Bedrock. GovCloud is purpose-built to meet government regulatory, security isolation, and compliance requirements.

The broader context is equally notable: in Q2 FY2026, AWS operating income grew 63% year-over-year and revenue grew 37%, giving it the financial confidence to continue investing in AI infrastructure. AI adoption in the government market is accelerating, and platforms that can satisfy both compliance requirements and model diversity are well-positioned to lead in this segment.

On a related note, the U.S. Department of Defense has officially launched AI services for joint forces on the GinAI.mil platform, onboarding over 1.7 million users in 9 months. The platform supports Impact Level 5 controlled unclassified information and offers deep reasoning capabilities across Auto, Fast, and Expert modes. Large-scale deployment of government AI infrastructure has moved from concept to reality.

WeChat Pay AI Dedicated Card: Completing the Loop from Conversation to Payment

WeChat Pay officially announced the launch of an AI Dedicated Card, which — following integrations with WorkBuddy and QCloud — now also supports DeepSeek Harness and OpenCloud authorization. Users can complete the full purchase journey, from AI-powered recommendations to checkout, directly within a conversation, and can pay to invoke over 700 PaySkills available on SkillHub.

The card's design reflects clear security considerations: it is isolated from the main account, has a user-defined spending limit, and requires the user's final authorization for each transaction. This "controlled delegation" model retains the convenience of AI automation while avoiding the financial risks of full delegation — a pragmatic implementation of AI Agent integration with financial services.

OpenMAICV 1.0 Tops GitHub Trending: Open-Source AI Classroom Draws Attention

The open-source multi-agent AI classroom project OpenMAICV 1.0 topped GitHub Trending within days of its release, surpassing 25,700 stars. The project supports one-click generation of complete interactive courses, and ModularAI simultaneously launched VoxCPM2 for voice cloning functionality within the platform.

OpenMAICV's viral rise reflects the current gap in AI education tooling: the combination of automated course generation, multi-agent collaborative teaching, and voice cloning is redefining content production efficiency in online education. For edtech practitioners, it's both an open-source benchmark worth tracking and a potential competitive reference point.

Perplexity Hybrid Mode: Local and Cloud Collaboration for Privacy-First AI

Perplexity is preparing to launch a Hybrid Mode on Mac, orchestrated by cloud-based Computer that delegates select tasks to local models — currently supporting Gemma 4, Quantum 3.6, and Perplexity's own models.

The standout feature is a built-in PrivacyGate micro-model that checks and intercepts personally sensitive information before it's uploaded to the cloud. This design philosophy — dynamically balancing cloud capability with local privacy — represents an innovative approach to privacy compliance pressure in AI applications, and is particularly appealing to enterprise users.

EU Brings ChatGPT Under DSA Very Large Platform Regulation

On August 31, the European Commission formally designated ChatGPT, Reddit, and Roblox as regulated entities under the Digital Services Act (DSA), based on their having more than 45 million monthly active users in the EU. ChatGPT was designated as a "very large online search engine," while Reddit and Roblox were designated as "very large online platforms."

The affected companies have a four-month transition period (through end of December 2026) to meet additional compliance requirements including risk assessments, content moderation, and transparency reporting. This marks the first time an AI product has been brought into the EU's major regulatory framework in its capacity as a "search engine" — a development whose compliance costs and operational implications should not be underestimated, and one that will set a precedent for other AI products' European market strategies.

Zhipu Semi-Annual Report: Revenue Up Nearly 400%, Losses Narrowing

At its H1 FY2026 earnings briefing, Zhipu disclosed total operating revenue of ¥954 million, up 399.7% year-over-year; scaled net loss was -¥2.071 billion, though losses narrowed by 12.1% compared to the same period last year.

Zhipu also revealed that its next-generation large model will follow a "large foundation model" approach, with specific parameters not yet disclosed. The plan involves continuous training and post-training optimization on top of the base model, with the goal of supporting full-scale business deployment from day one of release. The near-400% revenue growth combined with gradually narrowing losses reflects a growth curve typical of China's leading AI companies: scale first, with a path to profitability becoming increasingly clear.

Industry Outlook: Structural Shifts Across Multiple Fronts

This week's news covers a complete cross-section of the industry — from open-source models and commercial monetization to government procurement, privacy compliance, and tightening regulation. Several threads are worth watching closely:

• Open-source acceleration: DeepSeek's entry into multimodal and OpenMAICV's viral rise signal rapid expansion of the open-source ecosystem's capability frontier.
• Monetization maturity: ChatGPT ads breaking $1 billion in 200 days shows that AI product monetization pathways are diversifying.
• Deepening regulation: The EU DSA's classification of an AI search engine as a very large platform means compliance costs will become an intrinsic variable in product design.
• Localization trends: Perplexity's Hybrid Mode and the account isolation design of the AI Dedicated Card reflect how user demand for data control is driving product evolution.

These trends are not isolated — together, they sketch the structural characteristics of an AI industry transitioning from a "technology breakthrough phase" to a "scaled deployment phase."