Why This Moment Matters

In just one year, OpenAI launched several groundbreaking models, secured deals with major media and platform giants, refined its safety and governance structures, and saw its revenue soar into the billions. For developers, businesses, and curious minds alike, this represents a significant moment in which cutting-edge AI technology is transitioning from experimental demonstrations to practical, everyday applications.

Key Product Launches and Updates

OpenAI rolled out a variety of model and product updates in 2024, focusing on multimodality, reasoning capabilities, and enhanced developer usability.

GPT-4o: Real-Time, Multimodal AI

In May 2024, OpenAI unveiled GPT-4o, a multimodal model designed for real-time interactions in text, audio, and visual formats. This update included a new desktop application and more natural voice interactions, positioning GPT-4o as an advanced assistant capable of seeing, hearing, and speaking almost instantaneously (OpenAI).

Sora: Text-to-Video Generation

Earlier in the year, OpenAI launched Sora, a model adept at generating high-fidelity, minute-long videos from textual prompts (OpenAI). Though its release has been cautious with limited access, Sora reflects OpenAI’s bold ambitions in creative processes such as storyboarding, advertising, and synthetic b-roll creation.

Reasoning Models: o1 and o1-mini

By the end of 2024, OpenAI introduced a new range of models focused on structured reasoning, called o1 and o1-mini. These models prioritize step-by-step reasoning and methodical problem decomposition, delivering superior performance on complex math and coding tasks compared to previous GPT-4-class versions (OpenAI).

Cost-Effective Options: GPT-4o Mini

To expand accessibility and cater to budget-conscious applications, OpenAI launched GPT-4o mini, a smaller, faster, and more affordable model variant, drawing attention for its applications in chatbots, lightweight agents, and high-volume inference (Reuters).

GPT Store and Custom GPTs

In early 2024, OpenAI broadly launched its GPT Store, empowering users and teams to create and share specialized GPTs tailored to specific tasks. This marketplace approach is designed to enable non-experts to customize their assistants without requiring any coding expertise (OpenAI).

Voice Controversy and Product Adjustments

Alongside the new voice features in GPT-4o, OpenAI encountered public backlash regarding a voice option named Sky, which actor Scarlett Johansson claimed resembled her voice. After she expressed concerns about the unauthorized use of her voice, OpenAI quickly paused the Sky voice feature (Reuters, The Verge).

Strategic Partnerships: Apple, Media Licensing, and Data Deals

In 2024, OpenAI adopted a dual strategy focused on integration with major platforms and expansive licensing agreements to enhance model quality and ensure legal clarity.

Apple Integration

During WWDC 2024, Apple announced that ChatGPT would be integrated within Apple Intelligence for specific queries. This integration emphasizes privacy protections, including on-device handling when feasible, and allows for optional, user-approved interactions with ChatGPT without requiring separate accounts (Apple). OpenAI framed this integration as part of a broader strategy, indicating it won’t be exclusive.

Publisher and Platform Licensing

• **News Corp:** A multi-year agreement grants OpenAI access to a wide range of news content from outlets like The Wall Street Journal and the New York Post, allowing for proper attribution in display of content (News Corp).
• **Financial Times:** Announced in April 2024, this partnership permits the use of FT content with proper attribution in training and product experiences (Reuters).
• **Vox Media and The Atlantic:** In May 2024, OpenAI finalized licensing deals to access content while providing citations for these outlets, marking a significant step toward formalizing content access (The Verge).
• **Reddit:** Through a data licensing and partnership deal, OpenAI incorporated real-time community content into ChatGPT, offering new tools for Reddit moderators and users (Reddit).
• **Associated Press:** OpenAI’s ongoing agreement with AP from 2023 continued into 2024, enabling access to part of AP’s text archive and allowing experimentation with OpenAI’s tools (OpenAI).

Overall, these licensing agreements aim to mitigate legal risks, enhance factual accuracy, and ensure proper attribution. They also highlight how major AI players are renegotiating the value exchange with publishers in the generative era.

Legal and Regulatory Landscape

The rapid development of OpenAI’s products has taken place against a backdrop of heightened regulatory and legal scrutiny, especially concerning copyright and safety issues.

Copyright Litigation

In late 2023, The New York Times filed a lawsuit against OpenAI and Microsoft, alleging copyright infringement related to the training data and the reproduction of its content. By July 2024, a U.S. judge allowed significant parts of this case to progress after narrowing other claims, marking it as a crucial moment for the future of fair use and data licensing norms in the generative AI space (Reuters).

Privacy and Safety Standards

Regulatory bodies in both the U.S. and Europe have expressed increasing concern over how AI models are trained, assessed, and deployed. In Europe, the AI Act advanced toward implementation in 2024, while national authorities investigated various aspects of generative AI services related to transparency and user data handling. OpenAI has begun publishing more detailed model specifications and safety documentation, although the frequency and depth of the information provided varies.

Safety, Governance, and Leadership Changes

Discussions about AI safety and governance intensified in 2024, coinciding with significant leadership changes within OpenAI’s research team.

Leadership Departures and Promotions

In May 2024, co-founder and chief scientist Ilya Sutskever left OpenAI during a broader reshuffle within the research organization. Jakub Pachocki, a long-standing leader in research, was appointed as the new chief scientist (The Verge).

Also in May, Jan Leike, a key safety leader focused on long-term risk research, resigned, publicly criticizing the company’s prioritization of safety spending and processes. His departure highlighted ongoing debates regarding the balance between rapid deployment and long-term risk management (Reuters).

Superalignment Team Changes and New Committee

Reports in May 2024 indicated that OpenAI’s Superalignment team was restructured and, in practice, disbanded as a distinct entity, with safety efforts integrated across the organization (Wired). Shortly thereafter, OpenAI announced the formation of a Safety and Security Committee, led by board members and senior executives, aimed at reviewing and enhancing safety procedures for forthcoming model releases (The Verge).

Implications for Users and Enterprises

• Anticipate more detailed safety documentation and usage policies linked to major releases.
• Enterprises should review emerging safety features such as enhanced content filters and rate limits, aligning them with internal risk management protocols.
• Model behavior may change more frequently as safety procedures and governance checkpoints are updated.

Revenue and Business Model

OpenAI experienced rapid revenue growth throughout 2024, propelled by API usage, enterprise subscriptions, and partnership integrations.

Run-Rate and Growth

Multiple sources reported that OpenAI’s annualized revenue run-rate surpassed $3 billion by late 2024, illustrating strong enterprise demand for GPT-4-class models and new developer applications (Reuters).

Revenue Streams

• Enterprise and developer subscriptions for ChatGPT and the API, including platform credits, fine-tuning, and embeddings.
• Partnership income from integrations with platform partners and licensing agreements.
• Emerging monetization avenues via a marketplace model for custom GPTs tailored for teams and businesses.

Costs and Margins

Despite robust demand, the high computational intensity of model inference has influenced pricing strategies, prompting the introduction of smaller, more affordable models and usage-based billing. Furthermore, the relationship with Microsoft’s Azure infrastructure remains crucial for scaling operations (Microsoft).

Compute, Chips, and Scalability

The AI sector’s ability to train and deliver cutting-edge models depends heavily on computing resources. In 2024, OpenAI explored strategies for securing long-term access to advanced chips and data center capacities and engaged in ambitious discussions focused on expanding semiconductor supply (Reuters). Ultimately, OpenAI’s future plans hinge on having reliable and cost-effective infrastructure at a significant scale.

Competition and the Broader Market

OpenAI’s position in 2024 was further defined against a backdrop of rapidly evolving competition.

• **Anthropic:** The Claude 3 family (and later Claude 3.5 Sonnet) achieved notable benchmarks in reasoning and coding, with a strong focus on safety and user-friendliness (Anthropic, Anthropic).
• **Google:** Gemini 1.5 advanced long-context reasoning and multimodal capabilities, catering specifically to developers needing extensive context windows (Google).
• **Meta:** Llama 3 enhanced the open model ecosystem and stimulated a surge in fine-tuned domain models and edge deployments (Meta).

For buyers, the market has evolved into a portfolio decision: select a primary model family, maintain backups for specialized tasks, and test various specialized or open-weight models depending on privacy or latency needs.

What This Means for Developers and Businesses

For teams assessing OpenAI’s offerings, several actionable themes emerged in 2024.

Practical Guidance

• **Model Selection:** Use GPT-4o or o1-class models for complex tasks and opt for smaller variants (like GPT-4o mini) for high-volume or latency-sensitive applications.
• **Grounding and Retrieval:** Supplement generation with retrieval methods and structured tool-use to enhance reliability and minimize inaccuracies.
• **Audit and Safety:** Regularly review usage policies, content filters, and rate limits, implementing tiered safeguards for critical actions (e.g., additional checks before executing code or making external calls).
• **Data Strategy:** Where feasible, leverage licensed or first-party data. For sensitive tasks, evaluate data retention and privacy protocols with your provider.
• **Vendor Resilience:** Establish abstractions to ensure flexibility in switching model providers or versions without overhauling your entire stack.

Emerging Use Cases

• Customer support and sales assistants leveraging tools for CRM and invoicing.
• Code assistants for tasks such as refactoring, generating tests, and facilitating migrations.
• Knowledge agents interfacing with enterprise data through secure retrieval.
• Creative workflows in marketing, product design, and video prototyping utilizing models like Sora.
• Analytics copilot tools that convert natural language into SQL queries and BI dashboards.

Risks and Open Questions

Despite advancements in OpenAI’s technology, several uncertainties persist:

• **Copyright Outcomes:** Legal decisions may impact how training data is licensed and how outputs are credited.
• **Safety Governance:** The independence and transparency of external oversight for advanced models remains uncertain.
• **Cost and Availability:** Computational limitations could influence pricing, latency, and regional service offerings.
• **Platform Dependence:** Heavy reliance on a single cloud or chip supplier could introduce concentration risks.
• **Regulatory Divergence:** Different regional regulations may necessitate product adaptations in various markets.

The Bottom Line

OpenAI embarked on 2024 as the leader in generative AI and concluded the year with deeper platform integrations, a broader model portfolio, billions in annualized revenue, and increased scrutiny. For professionals and teams, the key takeaway is to view AI capabilities as dynamic: continuously pilot innovations, measure results, and remain adaptable to changes in models and licensing policies.

FAQs

Is OpenAI still the state-of-the-art for general-purpose AI?

OpenAI remains a leader, especially in multimodal interactions and overall usability. However, competitors like Anthropic, Google, and open-weight models from Meta are making strides in specific areas. The choice ultimately depends on your use case, data, and constraints.

What changed with safety in 2024?

OpenAI restructured its long-term risk research team, underwent leadership changes, and established a Safety and Security Committee to oversee forthcoming releases. Although public documentation and usage policies have expanded, there are ongoing calls for greater transparency and external oversight.

How does the Apple partnership affect developers?

The partnership enhances distribution for ChatGPT features within Apple environments, promoting privacy-first approaches to AI queries. While this doesn’t replace direct API access for custom applications, it may raise user expectations regarding latency, privacy, and reliability.

How should I choose between o1, GPT-4o, and smaller models?

Select the model based on the task: use o1 for intricate reasoning (math, coding, planning), GPT-4o for multimodal conversational tasks, and smaller models (like GPT-4o mini) for high-throughput or budget-sensitive applications. Benchmark using your actual data and workflows.

What is the outlook for 2025?

Expect advancements in faster, more reliable multimodal agents with enhanced enterprise controls and ongoing consolidation of content licensing. Legal outcomes regarding copyright could significantly impact training data practices and product design.

Sources

1. Introducing GPT-4o – OpenAI, May 2024
2. Sora – OpenAI, Feb 2024
3. OpenAI o1 – OpenAI, Sept 2024
4. OpenAI unveils cheaper, faster GPT-4o mini – Reuters, July 2024
5. Introducing the GPT Store – OpenAI, Jan 2024
6. Scarlett Johansson says OpenAI imitated her voice – Reuters, May 2024
7. OpenAI pauses Sky voice after Johansson comparison – The Verge, May 2024
8. Introducing Apple Intelligence – Apple, June 2024
9. News Corp and OpenAI announce a strategic content agreement – News Corp, May 2024
10. FT partners with OpenAI – Reuters, April 2024
11. OpenAI signs licensing deals with The Atlantic and Vox Media – The Verge, May 2024
12. Reddit and OpenAI partner – Reddit, May 2024
13. Partnership with The Associated Press – OpenAI, July 2023
14. Judge allows key NYT lawsuit claims to proceed – Reuters, July 2024
15. Ilya Sutskever leaves OpenAI – The Verge, May 2024
16. OpenAI’s Jan Leike resigns over safety concerns – Reuters, May 2024
17. OpenAI’s Superalignment team reportedly dissolved – Wired, May 2024
18. OpenAI forms Safety and Security Committee – The Verge, May 2024
19. OpenAI annualized revenue at $3.4 billion – Reuters, Sept 2024
20. Microsoft and OpenAI: Our partnership – Microsoft, Nov 2023
21. Introducing Claude 3 – Anthropic, Mar 2024
22. Claude 3.5 Sonnet – Anthropic, June 2024
23. Gemini 1.5 Pro with 1M+ context – Google, Feb 2024
24. Meta Llama 3 – Meta, Apr 2024
25. Sam Altman explores funding for AI chips – Reuters, Feb 2024