Is the Era of Gemini, Claude, and ChatGPT Over?... AI Now Competes on Cost-Effectiveness [Weekend Money]

Latecomer Models Gain Ground with Price and Efficiency
Cost Reduction through 'Routing' Instead of Single LLMs

The monopoly of cutting-edge (frontier) artificial intelligence (AI) models that had been leading the global AI market is being broken, with latecomers reshaping the landscape by leveraging overwhelming price competitiveness and efficiency, according to recent analysis.

Recently, Ilhyuk Kim, a researcher at KB Securities, stated, "The scarcity value of leading models such as OpenAI's ChatGPT, Anthropic's Claude, and Google's Gemini is diminishing." He explained that it has become difficult for frontier model companies to maintain high prices based solely on technological prowess, and that corporate clients are reducing costs by selectively using various AI models according to the nature of their work.

Previously, on July 9, SpaceX re-entered the AI model competition by unveiling Grok 4.5. Kim noted, "In various benchmark evaluations, it has demonstrated performance comparable to Claude Fable 5, Claude Opus 4.8, and OpenAI GPT 5.5."

The key point is price competitiveness. Grok 4.5 is priced at $6 per 1 million output tokens, a $3.5 increase from the previous model, but its token efficiency has also been enhanced, resulting in lower cost burdens compared to competing models. Kim added, "As SpaceX, having secured funding through its initial public offering (IPO), intensifies its price-based offensive, OpenAI and Anthropic are facing mounting pressure to lower their token prices."

On June 30, Meituan, a Chinese life services e-commerce platform, released its AI model LongCat 2.0. Kim pointed out, "It is becoming common for companies to achieve enhancements in model performance through methods such as reinforcement learning and inference optimization." He also noted, "The fact that these improvements can now be achieved using Chinese-made accelerators, instead of U.S.-made ones, is another factor diminishing the scarcity value of frontier AI models."

The emergence of 'orchestration' technologies, which combine multiple models to achieve optimal efficiency rather than relying on a single AI model, is also putting pressure on frontier companies. On June 22, Sakana AI, a Japanese company, released Fugu, which is not an independent model but a multi-agent orchestration system that selects, validates, and integrates various large language models (LLMs) depending on the situation. When a user makes a single request, the system automatically selects the optimal models, assigns them roles, and generates a response internally.

Kim added, "Companies are actively utilizing 'routing,' which involves choosing different models according to the nature of their work, to lower the cost of adopting AI." He continued, "Orchestration capabilities are likely to evolve in a direction that achieves the best performance with smaller, cheaper models. As a result, large frontier models may be relegated to the role of overall task planning only."

This content was produced with the assistance of AI translation services.