Spring 2025 AI Model Usage Trends: Poe Platform Insights

The AI landscape is evolving at an unprecedented pace, and understanding demand and usage patterns remains a significant challenge. While standardized benchmarks and leaderboard platforms provide valuable insights, they often don't capture the dynamic shifts in model popularity. Poe, with its provider-agnostic access to the latest frontier models, offers a unique perspective on these trends.

This report analyzes weekly aggregated usage data from January 2025 to May 2025 among Poe users across several key domains: text, reasoning, image, video, and audio. Here’s what we found:

Frontier Labs Release Smarter General-Purpose Text Models

• OpenAI's GPT-4.1 Family: OpenAI’s GPT-4.1 family has seen a significant uptick in usage, with message shares increasing to around 10% within weeks of launch. This is particularly notable for its improved performance on real-world coding tasks.
• Google’s Gemini 2.5 Pro: Google's Gemini 2.5 Pro also showed strong growth, reaching message shares of about 5%. Like GPT-4.1, it excels in practical applications such as coding.
• Anthropic’s Claude Family: Anthropic’s Claude family, including models like Claude 3.5 Sonnet and Claude 3.7 Sonnet, saw a significant decline in usage, with an absolute drop of around 10% over the same period.
• DeepSeek R1: DeepSeek's viral moment appears to have waned as other affordable, verbose reasoning models have been released. DeepSeek R1’s message share declined from a peak of 7% in mid-February to 3% by the end of April.

Cannibalization Within Provider Offerings

Similar to findings in our previous report, new flagship models within an individual provider's offering often cannibalize their predecessors. For example:

• Claude-3.7-Sonnet Over Claude-3.5-Sonnet: Poe subscribers rapidly adopted Claude 3.7 Sonnet over Claude 3.5 Sonnet. However, the latter retains a significant user base due to its reliability and familiarity.

Image and Video Generation Become Increasingly Competitive

• Image Generation: The image generation domain is becoming highly competitive with multiple models vying for dominance. Notably, models like Imagen-2.0 have gained significant traction, driven by their ability to produce high-quality images with minimal latency.
• Video Generation: The video generation space is also seeing rapid innovation. Models like Imagen Video are leading the charge, offering realistic and coherent video sequences that can be generated in real-time.

Early Signs of Diversification in Audio

• Audio Generation: While still a nascent field, audio generation models are beginning to show promise. Poe users are experimenting with various models for tasks ranging from music composition to voice synthesis. Early indicators suggest a growing interest in this domain as more sophisticated models become available.

Key Takeaways

1. Rapid Model Iteration: The speed at which new models are released and adopted is unprecedented. This highlights the importance of staying updated with the latest advancements.
2. User Preferences Shift Quickly: User preferences can shift rapidly, especially when new models offer significant improvements in specific areas like coding or reasoning.
3. Competitive Domains: Image and video generation are becoming increasingly competitive, driven by advances in model architecture and optimization techniques.
4. Diversification in Audio: The audio domain is showing early signs of diversification, with more users exploring the capabilities of AI-generated content.

We hope these findings provide researchers and practitioners with valuable insights into the rapidly evolving AI ecosystem. Stay tuned for our next report, where we will delve deeper into these trends and explore new developments.