Who gets the Mandate of Heaven?

The dynamics of the LLM arms race in one slide. ***Arms Race in the Large Language Model Market (LLM) It is simple: everyone tries to get the highest accuracy at the lowest price. And the “frontier” reflects the best options for combining these two parameters at the moment. *The chart shows [1]Different versions of language models (OpenAI, Deepseek, Google "Gemini", Anthropic, etc.) Relative to: cost (x-axis): price per million tokens - the more right the point, the cheaper the model (below the cost per million tokens). quality (y-axis)LMSys Elo rating - the higher the point, the stronger the model (Better quality of responses/results).

На диаграмме видны две основные "границы эффективности" (pareto frontier): 
•  Синяя линия от OpenAI, показывающая их модели
•  Оранжевая линия от Gemini 2, которая, судя по надписи, предлагает "лучше, дешевле, круче"
•  Более дорогие и мощные модели в верхней левой части (например, различные версии GPT-4) 
•  Средний сегмент в центре (Claude 3.5, Gemini 1.5) 
•  Более доступные модели в правой части (Amazon Nova Lite, Gemini 1.5 Flash)
```**


Key findings** (as of February 2025)
• Champion in price-performance ratio - Gemini 2.0 Flash Thinking (Better than DeepSeek R1 (elo) cheaper
• The cost of GPT's capabilities4 fell 1000 time 18 months 
• The speed of growth of the possibilities of models is simply unthinkable - it does not happen ... but it is!*

Google DeepMind experts believe they are close to getting a Mandate of Heaven ("Mandate of Heaven" (, Tianming)) *[\[2\]](https://x.com/swyx/status/1887263009921507656)*. When it is said that the company has a “Mandate of Heaven” in AI, it means that it is in a leading position not only because of market share, but also because of the recognition of its technological superiority and innovative leadership.

But competitors are unlikely to agree.* 😊

#ИИгонка