As generative AI continues its rapid evolution, Poe—a leading platform aggregating access to over 100 AI models—released its Spring 2025 AI Model Usage Trend Report, providing a comprehensive analysis of user preferences and shifting dynamics across multiple AI modalities. Drawing from millions of user interactions between January and April 2025, the report offers valuable insights into which AI models are gaining traction and which are losing ground in text generation, reasoning, image, video, and audio domains.
Surge in Reasoning Models: Gemini 2.5 Pro and GPT-4o Lead the Pack
One of the most striking trends is the rapid rise of reasoning-capable AI models. Usage of such models grew sharply from about 2% in January to 10% by April. Google’s Gemini 2.5 Pro quickly emerged as a dominant player, capturing approximately 31% of the reasoning model market within six weeks of its release. OpenAI’s GPT-4o further expanded its influence, commanding 35.8% of all message activity on Poe, with its GPT-4.1 variant contributing an additional 9.4%.
Other models such as OpenAI’s o1-pro, o3, and o4-mini also gained rapid adoption, while DeepSeek R1, which peaked at 7% usage in February, saw a decline to 3% by late April. Emerging hybrid reasoning models like Gemini 2.5 Flash Preview and Qwen3 demonstrated potential adaptability but remain niche with combined usage around 1%.
This surge reflects a maturing AI market where users increasingly demand models capable of complex, multi-step reasoning rather than simple text generation, pushing providers to innovate on reasoning capabilities to maintain competitiveness.
Text Generation: OpenAI and Google Dominate, Claude Declines
In the broader text generation category, OpenAI and Google maintain a strong dual dominance. GPT-4o leads with a 35.8% usage share, while the newly launched GPT-4.1 family quickly climbed to 9.4% within weeks, signaling strong user demand for cutting-edge performance. Google’s Gemini 2.5 Pro also captured around 5% shortly after debut.
Conversely, Anthropic’s Claude series experienced a notable 10% decline in usage share over the same period. While Claude 3.7 Sonnet overtook Claude 3.5 Sonnet in user preference, the latter still holds a stable 12% share. This shift underscores the competitive pressure Claude faces amid rapid innovation from OpenAI and Google.
Image and Video Generation: Rising Stars and Shifting Leadership
In image generation, Google’s Imagen3 steadily increased its market share from roughly 10% to 30% during early 2025, challenging the longstanding leader Black Forest Labs’ FLUX models, whose share declined from 45% to 35%. OpenAI’s GPT-Image-1 also made a swift impact, reaching 17% usage within just two weeks of release.
The video generation sector witnessed a dramatic rise of China’s Kuaishou “Kling” model family. Kling 2.0 Master, launched in late April, captured approximately 21% of the video generation market within three weeks, surpassing competitors like Runway and Google’s Veo2. Since its initial release, Kling AI has amassed over 22 million global users, with monthly active users growing 25-fold, and users generating over 168 million videos and 344 million images cumulatively. Kling’s leadership in dynamic quality, semantic responsiveness, and visual aesthetics marks it as a key global contender in AI-driven video creation.
Audio Generation: ElevenLabs Maintains Stronghold
In the audio domain, particularly text-to-speech (TTS), ElevenLabs continues to dominate with around 80% usage share. While innovative newcomers like Cartesia and Unreal Speech show promise, they have yet to achieve widespread adoption. The robust performance of ElevenLabs highlights the growing importance of high-quality, natural-sounding voice synthesis in applications such as audiobooks, podcasts, and virtual assistants.
Market Dynamics and Emerging Patterns
The report highlights a clear reshaping of the AI model landscape. Models that once enjoyed popularity, such as DeepSeek R1, have seen usage decline amid the influx of more affordable and capable reasoning models. The rapid adoption of OpenAI’s latest releases and Google’s Gemini series underscores the market’s preference for models offering superior reasoning and multi-step task capabilities.
Additionally, hybrid reasoning models that dynamically adjust reasoning depth during interactions are emerging, though they currently represent a small fraction of usage. This points to a future where AI systems become more adaptive and context-aware.
Limitations and Context of the Report
It is important to note that Poe’s report reflects usage patterns within its own platform, which primarily serves consumer-oriented, multi-modal AI applications. As such, the data may not fully represent enterprise deployments, open-source AI tools, or region-specific platforms. Nonetheless, the report serves as a valuable real-time window into user preferences and innovation trends in the rapidly evolving AI ecosystem.
Outlook: Toward a More Intelligent and Diverse AI Ecosystem
Looking ahead, the continued rise of reasoning models signals a shift toward AI systems capable of handling increasingly complex tasks with greater precision. The growth in image and video generation models, especially those with strong semantic understanding and aesthetic quality, suggests expanding creative applications.
Audio AI’s maturation further complements the multi-modal AI landscape, enabling richer, more immersive user experiences. Providers will need to balance innovation, cost, and accessibility to capture and sustain user engagement in this competitive environment.
In summary, Poe’s Spring 2025 AI Model Usage Trend Report reveals a dynamic and rapidly shifting AI landscape. Leading tech giants OpenAI and Google consolidate their dominance in text and reasoning models, while emerging players like Kuaishou’s Kling reshape the video generation market. Audio synthesis remains a specialized yet vital domain, with ElevenLabs at the forefront. As AI technology continues to mature, user preferences increasingly favor models with advanced reasoning and multi-modal capabilities, setting the stage for the next wave of AI-driven innovation.