It was just a few months ago in May when we hailed the launch of GPT-4o, which brought groundbreaking improvements in speed and multi-modality across text, audio, and images.
Since then, various models have quickly caught up. Anthropic’s Claude 3.5 Sonnet launched in June, achieving the highest score on HumanEval, a benchmark for coding ability. Meta launched Llama 3.1 in July, leading the Grade School Math 8K (GSM8K) and AI2 Reasoning Challenge (ARC) benchmarks, which measure mathematical reasoning and question-answering capabilities.
In our testing at Cerulean, we’ve found little to distinguish between the leading models. So, what happens when products become commoditized? Prices must come down. While consumers and enterprises continue to pay $20-25 per month, the cost for developers using these models via API is dramatically decreasing.
Market consolidation.
As leading model developers cut prices to attract developers, several well-funded competitors have exited the field::
Inflection AI: Raised $1.5B, valued at $4B, acqui-hired by Microsoft in March 2024. Its chatbot, Pi, never exceeded 1M users.
Adept AI: Raised $400M, valued at $1B, acqui-hired by Amazon in June 2024. Adept aimed to automate tasks by interacting with software on users’ computers.
Character AI: Raised $150M, valued at $2.5B, acqui-hired by Google two weeks ago. It struggled to acquire paying users despite offering custom chatbots.
These companies raised significant capital by most standards, but it wasn’t enough relative to the high costs of training and maintaining these models. OpenAI mentioned that GPT-4 cost “more than $100M” to train, while Llama 3.1 is estimated to have cost Meta hundreds of millions. Mark Zuckerberg has indicated that training Llama 4.0 will require nearly ten times more compute power compared to Llama 3.
Few companies have the scale to justify these investments, so only a select few can survive. The leading model makers—OpenAI (Microsoft), Anthropic (Amazon), Gemini (Google), and Llama (Meta)—are each backed by tech giants. These leaders will likely compete more on price as capabilities converge.
Focus on practical applications.
As foundational models converge and consolidate, the next wave of investment will focus on translating these advanced capabilities into solutions for real-world challenges.
The latest trend we’ve observed is major corporations investing in AI-driven tools for knowledge management, which involves organizing an organization’s collective information to improve decision-making speed and quality. By breaking down silos and making information readily available, AI solutions enable employees to move quickly without the hassle of time-consuming information searches. This use case is especially valuable for companies that have grown via acquisition, resulting in historical records being held in various legacy data stores.
Recent examples among large enterprises include:
Morgan Stanley: Developed AI-powered tools to help advisers quickly track down research team information.
TD Bank: Leveraged GPT-4 to develop an assistant that helps call center agents answer customer inquiries more effectively.
Amazon: Building an Anthropic-powered AI tool to optimize content management systems and boost employee productivity.
Additionally, the WSJ recently reported that Glean is in talks to raise $250M at a $4.5B valuation, doubling its valuation from six months ago. Glean offers a product that helps employees look up information spread across the company, with successful case studies including:
Duolingo: Introduced a conversational assistant to facilitate information sharing and cross-functional collaboration, saving $1.1M in costs associated with information searches.
Webflow: Implemented Glean to address siloed and fragmented knowledge, unlocking 300+ hours monthly from time spent searching for information.
Novel applications like knowledge management are building on long-standing use cases to increase operational efficiency and improve customer experiences. Companies like Starbucks and Netflix have long employed AI-based algorithms for personalized loyalty and content recommendations, while retailers like Walmart and H&M have used AI to optimize supply chains, increasing item availability by up to 20% and reducing labor costs by up to 15%.
Finally, news earlier this week from Ramp, a corporate expense management platform, reveals that average spending on AI vendors has risen by 375%.
Source: Eric Glyman, Ramp CEO
The insight here is clear: companies are moving beyond experimenting with GenAI tools (often done on corporate credit cards rather than through AP invoices). Leaders are finding ever more ways to use AI to benefit their customers and organizations, and select examples of these innovations are continuing to surface in the public domain.