80% to 25% reversal! Latest ranking of large model coding capabilities: Is Anthropic no longer the only god?

The OpenRouter website provides an interesting model usage ranking, categorized by different usage scenarios (such as programming, writing, etc.). Someone has organized the data in the "programming" category and grouped it by model providers, revealing some interesting market trends.

Data shows that from December 2024 to February 2025, the model from Anthropic had a traffic share of about 80% in programming scenarios, demonstrating a strong performance. This data also indirectly confirms the general observation at the time: developers indeed favored using Claude 3.5.

However, the market landscape changed afterward. With the release of Google Gemini 2.5 Pro and (to a lesser extent) GPT-4.1, these two models began to capture market share from Anthropic, causing its share to drop below 25% at one point.

Recently, Anthropic released the next generation Claude 4 model, particularly the Sonnet model, which helped its share rebound significantly. Nevertheless, its market share has yet to return to the previous peak level of about 80%.

Of course, it is important to note that OpenRouter's data does not fully represent the entire market, as its usage data may be skewed due to concentrated usage by a few large clients or occasional traffic spikes. However, this trend still indicates a fact: intense market competition is beneficial for users.

OpenRouter is an aggregator or router for large language models (LLM)

It does not develop its own models but serves as an intermediary platform that integrates numerous models from different top providers (such as OpenAI, Google, Anthropic, deepseek, Meta, etc.) behind a unified API interface.

Its core features and functions include:

Unified API interface: Developers only need to use OpenRouter's API to call models from different companies. Its API format is generally compatible with OpenAI's format, meaning that if you have written code to call GPT-4, you can almost seamlessly switch to calling Claude 3, Gemini Pro, or other models, significantly reducing development and switching costs Access to numerous models: With one account, you can use almost all mainstream and some niche AI models available on the market. This is very convenient for developers who need to compare the performance of different models or wish to select the most suitable model based on tasks.

Unified billing and cost management: You do not need to register separately with each model provider, apply for API Keys, and bind credit cards. All model usage will be billed uniformly on OpenRouter; you only need to recharge one account and can clearly see the expenses for different models.

Intelligent routing function (Auto-Routing): This is the origin of its name "router." OpenRouter provides a special "auto" model that can automatically determine and select the most cost-effective or best-performing model to process your requests based on your prompts, helping users achieve a balance between cost and effectiveness.

Model rankings and discovering new models: OpenRouter generates model rankings based on real user usage data. This not only reflects market trends but also helps developers discover some newly released or outstanding niche models.

Risk warning and disclaimer

The market has risks, and investment requires caution. This article does not constitute personal investment advice and does not take into account the specific investment goals, financial situation, or needs of individual users. Users should consider whether any opinions, views, or conclusions in this article align with their specific circumstances. Investing based on this is at your own risk