17-May-2025 :

Poe has published its Spring 2025 AI Model Usage Trends report, offering a deep dive into how users are engaging with text, reasoning, image, video, and audio models. The report reflects significant shifts in preference as newer AI models gain traction and established players see usage declines.
Among text-based models, GPT-4.1 and Gemini 2.5 Pro gained 10% and 5% of total message volume shortly after their launch, respectively. Meanwhile, Claude experienced a 10% decline over the same period, highlighting changing user preferences.
The biggest surge was seen in reasoning models, which grew from just 2% to 10% of all text messages since January. Notably, Gemini 2.5 Pro accounts for nearly a third of that subcategory, showcasing the appeal of more advanced, logic-capable systems.
Image generation usage also saw disruption. GPT-image-1 rose to 17% usage, making it a direct competitor to Black Forest Labs’ FLUX and Google’s Imagen3 models. In the video category, China’s Kling model rapidly captured ~30% share after its release, signaling demand for high-quality video generation tools.
Audio AI also remained a dominant category, with ElevenLabs retaining its lead with an 80% share of audio output usage, reinforcing its continued appeal in synthetic voice generation.
These findings show that user adoption patterns can shift dramatically in weeks, not months, and offer a real-world counterpoint to benchmark scores. Poe’s platform, which enables rapid side-by-side comparisons, reveals emerging trends in what users actually prefer when selecting AI models for diverse tasks.