For a while, OpenAI and ChatGPT seemed synonymous with cutting-edge Large Language Models (LLMs). They captured the public imagination and set the benchmark for AI assistants. However, the landscape is rapidly evolving, and a new wave of models and platforms are challenging OpenAI’s once-unquestioned dominance. Data from platforms like OpenRouter and the LM Arena Leaderboard suggest that the AI throne is no longer solely occupied by OpenAI.

OpenRouter Rankings: A Glimpse Beyond OpenAI

OpenRouter Rankings offers a unique perspective on LLM usage and preference. This platform acts as an aggregator, routing API requests to various LLM providers. Their weekly rankings, based on API usage, provide a real-world snapshot of which models developers are actively choosing. The platform’s data shows a dramatic surge in AI API usage, with completion requests growing from under 100B in early 2024 to over 800B by January 2025, highlighting the explosive growth in LLM adoption. This exponential increase, particularly steep in the latter half of 2024, suggests not just growing developer interest, but also the expanding integration of LLMs into production applications and services.

The OpenRouter Weekly Rankings highlight a significant trend: diversity. As of February 12, 2025, the top models in their weekly rankings are far from being solely dominated by OpenAI. The top 20 models showcase a wide array of providers: 

While OpenAI’s GPT-4o-mini is present, the list is heavily populated by models from Anthropic (Claude), Google (Gemini Flash & Pro), DeepSeek, Mistral, Meta (Llama), Nous, WizardLM, and Qwen. This demonstrates that developers are actively utilizing and exploring a wide range of LLMs beyond OpenAI’s offerings.

This diversification suggests several key shifts: First, high-performance LLMs are no longer exclusive to one company. Second, developers are actively seeking variety and potentially cost-effectiveness. OpenRouter’s platform facilitates access to different models, indicating a clear demand for choice beyond a single provider.

LM Arena Leaderboard: Community Voices Across Modalities

The LM Arena Leaderboard provides an even broader perspective, extending beyond text-based LLMs to include Text-to-Image and Vision models. This community-driven platform, based on pairwise evaluations, reveals user preferences across different AI modalities.

The LM Arena Leaderboard reinforces the trend of diversifying AI leadership. In Text-based LLMs, while ChatGPT-4o-latest and other OpenAI models remain competitive, they are closely matched and sometimes surpassed by models like Gemini-2.0-Flash-Thinking-Exp-01-21, DeepSeek-R1, and Qwen2.5-Max. This mirrors the OpenRouter data, showing strong competition in the core LLM space.

However, the shift is even more pronounced when we look at Text-to-Image and Vision categories on LM Arena. While OpenAI’s DALL-E models are present, they do not dominate the top spots. Instead, models from Midjourney, Ideogram, and other specialized providers frequently achieve higher rankings in image generation quality and user preference. Similarly, in Vision tasks, models beyond OpenAI are demonstrating strong capabilities and community favor.

This multi-modality perspective highlights that leadership in AI is becoming increasingly specialized. Different companies and open-source projects are excelling in specific areas, rather than one entity dominating across all modalities.

Beyond the Throne: A Multi-Polar AI World

The combined picture from OpenRouter and LM Arena paints a clear trend: the AI landscape is becoming increasingly multi-polar across all modalities. While OpenAI undoubtedly played a pioneering role and remains a major force in text-based LLMs, their sole dominance is being challenged, and in areas like image and vision, they face even stronger competition.

  • Emerging Contenders Across Modalities: Companies like Google, Anthropic, DeepSeek, Midjourney, Ideogram, and open-source communities are producing models that are not only competitive with OpenAI’s offerings in text, but often lead in emerging areas like image and vision generation.
  • Platform Diversification Extends to All AI: Platforms like OpenRouter, facilitating access to diverse text models, are mirrored by emerging platforms for image and vision AI, further empowering users with choice.
  • Specialization and Best-of-Breed Approaches: The rise of specialized models for text, image, and vision suggests a move towards “best-of-breed” AI solutions. Users are no longer limited to a single provider for all AI needs but can select the most effective model for each specific task.
  • Open Source Strength Across Modalities: The success of open-source models extends beyond text, with open initiatives making significant strides in image and vision AI as well, fostering innovation and accessibility.

Conclusion: A Healthy Competition Benefits Everyone

The idea of a single “king” in the rapidly evolving field of AI is likely becoming outdated. The data from OpenRouter and LM Arena strongly suggests a more dynamic and competitive ecosystem is emerging across text, image, and vision modalities. This is ultimately beneficial for everyone. Healthy competition drives innovation, pushes performance boundaries, and provides users with more diverse, powerful, and accessible AI tools. While OpenAI remains a crucial player, the rise of strong competitors, specialized models, and open platforms indicates a future where the AI throne is shared, and the real winners are the users who benefit from this exciting progress in the multi-polar AI world.

Links: