OpenAI GPT-5.2 Models Unveiled, Disney Invests $1 Billion

OpenAI has launched its new GPT-5.2 models, coinciding with the company's tenth anniversary. The release includes three distinct versions: GPT-5.2 Instant, optimized for response speed; GPT-5.2 Thinking, designed for deep reasoning; and GPT-5.2 Pro, a high-end version. These models aim to cater to a range of applications, from everyday conversations to complex scientific research.

According to OpenAI, GPT-5.2 demonstrates superior performance in explicit knowledge work tasks across 44 professions compared to industry professionals. GPT-5.2 Thinking, in particular, shows significant improvements over its predecessor, GPT-5.1 Thinking, in knowledge-based tasks, programming, scientific problems, mathematics, and abstract reasoning. The model achieved a perfect score in the AIME 2025 mathematics competition and either matched or surpassed 70.9% of human experts in OpenAI's professional work benchmark test, GDPval.

Fidji Simo, OpenAI's Chief Product Officer, stated that the three versions are intended to "unlock more economic value" by facilitating tasks such as spreadsheet creation, presentation writing, coding, image analysis, and long-text comprehension.

Disney Partnership and Industry Impact

In a separate development, Disney announced a $1 billion investment in OpenAI and a three-year partnership. This agreement will allow Sora and ChatGPT Images to generate content using over 200 characters from Disney, Marvel, Pixar, and Star Wars intellectual properties, excluding actors' likenesses or voices, starting early next year.

The collaboration means Sora will be able to create Disney characters, and some user-generated Sora videos will be featured on Disney's streaming platform, Disney+. Both companies plan to leverage OpenAI's models to develop new products and customer experiences, including features for Disney+ users. Disney also intends to deploy ChatGPT internally for its employees. Following the announcement, Disney's pre-market stock price saw an increase of over 2%.

This partnership marks a shift in Disney's AI strategy. Previously, Disney had opted not to allow its characters to appear in AI applications like Sora during copyright discussions with OpenAI. The move is viewed by industry observers as a significant collaboration between a leading AI company and a global storytelling giant, potentially offering new paradigms for AI-content platform cooperation.

Evolving Content-AI Collaboration Models

The Disney-OpenAI collaboration reflects a broader trend of content companies and AI developers forming partnerships. This "reconciliation of the century" follows a period where Hollywood studios, including Disney and Universal Pictures, had sued AI companies like Midjourney for alleged unauthorized use of copyrighted material to train AI models.

Current cooperation models between content providers and AI companies generally fall into three categories:

Selling Data: This involves content companies licensing their data to AI developers for model training. Examples include media outlets like Time magazine and the Associated Press providing their content to ChatGPT, and Visual China partnering with Kuaishou to train large models with its image library.
Building Models: Content providers collaborate with AI companies to develop customized models. Lionsgate, for instance, opened its library of 20,000 films to Runway, which will use these assets to build a new cinematic-grade model.
Building Ecosystems: This long-term approach involves both parties exploring future business and traffic opportunities. The Disney-OpenAI partnership is one example. Another is Suno's collaboration with Warner Music Group to explore AI-assisted music creation and new copyright revenue-sharing mechanisms. In China, ByteDance's Jiemeng AI partnered with Shanghai Film Group for AI animation using classic IPs.

These evolving models suggest that intellectual property and content remain crucial assets in the AI era.

Addressing AI Data Needs

The collaboration between Disney and OpenAI also addresses the increasing demand for high-quality data in AI development. For AI models to advance, relying solely on internet data is becoming insufficient. There is a growing need for structured, curated, and copyrighted data to achieve further "transformation."

Concerns about "AI data depletion" were raised by researchers at Epoch AI last year, who predicted that AI might exhaust available training data within approximately four years. Additionally, studies have indicated that AI models can suffer from reduced performance, or "brain damage," when exposed to excessive amounts of low-value, fragmented online content.

This demand for high-quality data has led to the emergence of companies like Mercor, which connect AI companies with freelancers who have professional backgrounds to "hand-teach" AI models with specialized knowledge. These freelancers, like actress Shao Yifan who taught AI how to consult, provide valuable, experience-based data. While Mercor has seen its valuation increase, it has also faced controversy regarding the nature of this work.

Direct content partnerships between AI companies and IP holders, such as the Disney-OpenAI agreement, offer another significant avenue for sourcing high-quality, compliant content, potentially driving further advancements in AI capabilities.

Sam Altman's Reflection on OpenAI's Tenth Anniversary

In a short essay reflecting on OpenAI's tenth anniversary, CEO Sam Altman noted the company's journey from an ambitious, almost impossible goal to its current position. He highlighted the "extraordinary optimism and joy" of the early days and the continuous effort to explore new challenges, from research directions to funding for computing power.

Altman pointed to foundational results achieved in 2017, including Dota 1v1 experiments pushing reinforcement learning, unsupervised sentiment neuron experiments demonstrating language model semantics, and human-preference-based reinforcement learning for AI alignment. He emphasized the need for powerful computing to scale these results.

The launch of ChatGPT three years ago and GPT-4 subsequently brought widespread attention, making Artificial General Intelligence (AGI) seem more attainable. Altman described the past three years as intense and full of responsibility, noting the rapid integration of the technology into the world. He expressed pride in the team's decisions and acknowledged his own responsibility for any missteps.

He also defended the iterative deployment strategy, which brought early versions of the technology to market to raise awareness and allow societal and technological co-evolution, calling it "one of the smartest decisions we made."

Altman concluded with optimism about OpenAI's future, stating that the company now has AI capable of outperforming many intelligent individuals in intellectual competitions. He believes that superintelligence is likely to be built within the next decade. He thanked users and customers for their support, which he said was crucial in bringing the technology from the lab to real-world application. Altman reiterated OpenAI's mission to ensure AGI benefits all humanity.