GPT-6 Imminent as OpenAI Pivots Strategy: Shuts Down Sora, Bets on Enterprise AI

The spotlight in the global artificial intelligence field is once again on OpenAI. The high-profile company has officially confirmed that its next-generation large model, GPT-6, codenamed "Spud," is set for release. However, accompanying this launch is a surprising decision: the complete shutdown of the once-viral video generation product, Sora. This series of actions marks a profound strategic shift for OpenAI at a critical juncture as it races toward a potential public listing.

According to OpenAI's official announcements and information shared by its management, the pre-training for GPT-6 has been completed at the Stargate data center in Texas. It is now in the final stages of safety alignment and API debugging, with a release expected as soon as within April. OpenAI President Greg Brockman confirmed GPT-6's existence in a recent podcast interview, emphasizing that "this is not an incremental improvement, but a major change in how we think about model development." CEO Sam Altman defined it in an internal memo as "a very powerful model that can truly accelerate economic development."

GPT-6: Architectural Innovation and Capability Leap

Information from industry channels suggests GPT-6's core upgrades break from the incremental path of previous models. It employs a novel "Symphony" architecture, achieving native unified processing of text, images, audio, and video for the first time, rather than the industry-common approach of stitching together multimodal modules. This means users can, within a single unified interface, generate front-end code directly from hand-drawn sketches or upload a video to deconstruct action details and generate corresponding scripts.

In tasks developers care most about—coding, reasoning, and AI agent performance—GPT-6 shows over a 40% performance improvement compared to its predecessor, GPT-5.4. Its context window has expanded from 1 million to 2 million tokens, capable of processing approximately 1.5 million words of text in a single instance. Public test data indicates its mathematical reasoning accuracy reaches 92.5%, with a code generation pass rate of 96.8%.

Shutting Down Sora: A Difficult Choice Under Commercial Realities

In stark contrast to the high anticipation for GPT-6 is the quiet exit of Sora. The shutdown of this video generation tool, hailed as a "blockbuster" by the industry just a year ago, reveals the severe challenges AI technology faces in commercial deployment.

According to estimates by Forbes magazine, the annual operating cost of the Sora project exceeded $5 billion (~36.4 billion yuan), while its total in-app revenue since launch was only about $2.1 million (~15.3 million yuan), completely unable to cover the high computing and operational costs. The computational power consumption for AI video generation grows exponentially, with the cost to generate a basic 10-second video clip around $1.3 (~9.5 yuan), and costs for complex scenes reaching up to $33 (~240 yuan).

To control losses, OpenAI continuously reduced users' free generation quotas, from an initial 30 clips per day down to 6, which directly led to a massive user exodus. Monitoring data from Appfigures shows Sora's 30-day user retention rate was a mere 1%, with the 60-day rate approaching zero.

Furthermore, copyright and compliance issues became another breaking point for Sora. Initially, Sora gained rapid popularity for its ability to generate content featuring Disney IPs and celebrity likenesses, but this also triggered a wave of copyright lawsuits and industry backlash. OpenAI was forced to tighten content generation rules from "available by default" to "requires explicit authorization," directly undermining the product's core appeal and leading to the termination of a potential $1 billion (~7.3 billion yuan) partnership with Disney.

Strategic Pivot: From Technological Exploration to Commercial Deployment

The shutdown of Sora and the full bet on GPT-6 represent, in essence, OpenAI's effort to reshape its commercial narrative on the eve of a potential IPO. Goldman Sachs estimates that while OpenAI's 2025 revenue exceeded $20 billion (~145.3 billion yuan), its losses still ranged from $14 billion to $15 billion (~101.7 billion to 109 billion yuan). To meet profitability expectations for a public listing, OpenAI must allocate its limited resources to the most commercially viable business.

Currently, enterprise services are the most certain "cash cow" in the AI industry. Competitor Anthropic's annual revenue exceeds $19 billion (~138 billion yuan), with about 80% coming from enterprise clients, presenting direct competitive pressure for OpenAI. OpenAI's strategy is shifting from attracting consumer users with flashy generative capabilities to serving business enterprises with stable, efficient productivity tools.

According to a report by The Information, there is internal disagreement at OpenAI regarding the timing of an IPO. Sam Altman hopes to go public as early as the fourth quarter of 2026, aiming to beat Anthropic, which may also IPO around the same time. However, Chief Financial Officer Sarah Friar believes the company won't be ready by 2026, citing concerns over a future $600 billion (~4.36 trillion yuan) server leasing commitment over five years and cash burn that far exceeds expectations.

Industry Diversification: The AI Race Enters Multiple Tracks

While OpenAI prepares intensively for the GPT-6 launch, other players in the global AI industry are not following its lead. Instead, they have提前 initiated a new round of differentiated布局, leading to a clear divergence in industry paths.

Company	Recent Move	Core Strategic Direction
Google	Launched the fully open-source Gemma4 series of large models on April 2.	Open-source ecosystem, capturing the on-device AI入口, freeing models from reliance on cloud computing power.
Alibaba	Launched Tongyi Qianwen Qwen3.6-plus, performing well on coding leaderboards.	Deep cultivation of vertical scenarios (e.g., programming), leveraging local场景 and supply chain advantages.
ByteDance	Doubao 2.0 supports private deployment to meet enterprise security and compliance needs.	Promoting the productization of AI assistants and enterprise services.
Tencent	Fully rolled out ClawBot plugin on WeChat, natively integrating an AI agent framework.	Deeply integrating AI capabilities into existing super-app ecosystems.
Microsoft	Launched the MAI series of in-house commercial models (voice, image).	Building an in-house AI model system to reduce dependence on external technology.
Anthropic	Completed pre-training for Claude Mythos, featuring recursive self-correction capability.	Focusing on model safety and capability breakthroughs, targeting the high-end enterprise market.

Google's Gemma4 is fully open-sourced under the Apache 2.0 license. Its 2-billion-parameter model can run offline on smartphones, yet its performance matches that of the previous generation's 27-billion-parameter model. Within 24 hours of release, total downloads surpassed 400 million, sparking a new frenzy in the open-source community. This move is the opposite of OpenAI's approach, aiming to build a broad developer ecosystem through open source and capture the future of on-device AI.

Chinese manufacturers, meanwhile, are accelerating deployment by leveraging their local场景 advantages. Alibaba's Tongyi Qianwen Qwen3.6-plus ranks second globally on the React专项 leaderboard, which tests complex web development capabilities. Tencent is natively integrating AI agent capabilities into WeChat, attempting to turn the chat box used by 1.2 billion users into an AI console.

Impact and Outlook: From Parameter Race to Deployment Practice

The arrival of GPT-6 seems to be shifting the industry's competitive focus from "can it be built?" to "can it be used?" Whether it's GPT-6's deep integration of agent capabilities, Google Gemma4's极致 optimization for on-device deployment, or Chinese manufacturers'深度打磨 in vertical scenarios, the essence is to move AI technology out of the lab and truly integrate it into specific work scenarios to solve real problems.

For Chinese large model manufacturers, this presents both challenges and opportunities. In recent years, domestic models have largely been in a catch-up phase. However, as the industry's inflection point shifts toward deployment and ecosystems, domestic players,凭借 their deep understanding of local industry pain points, vast user bases, and supply chain resources, have the potential to achieve differentiated breakthroughs. For example, in traditional industries like manufacturing, finance, and retail, demand for vertical AI solutions that address specific pain points is rapidly exploding.

For ordinary developers and users, there's no need for excessive anxiety over the disruption brought by the latest model versions. What's truly valuable is learning to transform AI tools into helpers that enhance work efficiency. The significance of AI lies in liberating people from repetitive and tedious tasks to engage in more creative activities.

With the imminent release of GPT-6, it may刷新 people's understanding of AI capabilities. But on the long road to the so-called "Artificial General Intelligence (AGI)" endpoint, the entire industry is collectively undergoing a more pragmatic修行: how to make technology create sustainable commercial value and truly empower thousands of industries.

Note: This article synthesizes information from OpenAI's official announcements, industry analysis reports, and reports from multiple technology media outlets.