A globally renowned AI model integration platform, Poe, has released the "2025 Spring AI Model Usage Trend Report," which delves into the demand and usage share changes of text, image, video, audio, and reasoning models based on millions of user interactions on the platform. The report reveals the rapid rise of breakthrough models such as OpenAI's GPT-4.1, Google's Gemini2.5Pro, and Kuaishou's Kling this year, as well as profound shifts in AI usage patterns. AIbase provides exclusive insights into the core highlights of this report and its far-reaching impact on the AI ecosystem.
Text Models: A Dual-Dominance between OpenAI and Anthropic
The Poe report shows that text generation remains the core area of AI usage, with OpenAI and Anthropic continuing to dominate. OpenAI's GPT-4o maintains a 35.8% usage share at the top, while the newly launched GPT-4.1 family quickly climbed to 9.4% within weeks, reflecting strong user demand for high-performance models. However, Anthropic's Claude series (such as Claude3.5Sonnet and Claude3.7Sonnet) saw a 10% decline in usage share during the same period, partly due to the rapid replacement of Claude3.7Sonnet for Claude3.5Sonnet, which still holds a stable 12% usage rate.
AIbase analyzes that the rapid iteration of new models is driving users to migrate from older models to newer ones, reflecting the "flagship effect" in the AI industry. Notably, DeepSeek's R1 model peaked at 7% in mid-February and dropped to 3% by the end of April, indicating that the influence of open-source models is weakening under the competition of high-performance closed-source models.
Reasoning Models: The Fastest-Growing Star Area
In early 2025, reasoning models became the fastest-growing area in AI usage, with their share increasing sharply from 2% to 10%, thanks to their high accuracy in complex tasks. Google's Gemini2.5Pro captured 31% of the reasoning model share within six weeks of its release, surpassing Anthropic's Claude series and becoming the leader in this field. OpenAI rapidly introduced o1-pro, o3-mini, o3, and o4-mini reasoning models, with users quickly migrating from o1 to o3 and o4-mini, demonstrating the flexibility of its ecosystem.
AIbase notes that xAI's Grok3, despite topping multiple problem-solving benchmarks upon its release in February, only accounts for 1% of reasoning usage. Its API currently supports only the reasoning functionality of Grok-3-mini. Emerging hybrid reasoning models (such as Gemini2.5Flash Preview and Qwen3) are further diversifying the field by adjusting reasoning levels through conversational tuning.
Image Generation: The Emergence of a Three-Power Struggle
The competition in the image generation field is becoming increasingly intense. Black Forest Labs' FLUX series remains in the lead with 35% market share at the end of April but has declined from 45% at the beginning of the year. Google's Imagen3 family steadily grew from 10% to 30%, keeping pace with FLUX. OpenAI's GPT-Image-1 launched via API in late April and reached 17% of the image generation share within two weeks, continuing the viral spread momentum of ChatGPT applications.
AIbase believes that this three-power struggle reflects the rapid maturation of the image generation market. User preferences for high-quality, low-cost models are driving Google and OpenAI's rapid catch-up, while FLUX's leadership still relies on its open-source ecosystem and community support.
Video Generation: Kling2.0 Emerges as a Contender
Due to high computational requirements, video generation usage is relatively limited. However, cost reductions and quality improvements in 2025 are driving its rapid growth. Kuaishou's Kling2.0-Master captured 21% of the video generation share within three weeks of its release in April, with the series reaching a total of 30%, surpassing Runway (31.6%) and Google's Veo-2 (39.8%). AIbase tests show that Kling2.0 stands out in dynamic quality and visual beauty, attracting early adopters from advertising, social media, and educational platforms.
AIbase predicts that as computational costs continue to decrease, video generation will experience explosive growth in the second half of 2025, with the rise of Chinese models like Kling adding new variables to global competition.
Audio Generation: ElevenLabs Remains Unchallenged
In the audio generation field, ElevenLabs continues to lead with an 80% usage share, offering diverse voice options and high-quality text-to-speech services. New competitors such as Cartesia, Unreal Speech, PlayAI, and Orpheus are emerging with unique sound effects and pricing models but have yet to challenge ElevenLabs' dominant position. AIbase believes that competition in audio generation will focus on voice personalization and low-latency optimization, with models like MiniMax's Speech-02 potentially challenging the existing landscape in the future.
Dynamic Reshaping of the AI Ecosystem
As a specialized media platform for the AI sector, AIbase highly commends Poe's 2025 Spring Report. The report reveals how breakthrough models like Gemini2.5Pro, GPT-4.1, and Kling2.0 are rapidly reshaping user preferences through performance improvements and cost optimizations. The explosive growth of reasoning models and internationalized competition in video generation signify the comprehensive expansion of AI applications from single-text generation to multimodal and complex tasks.
Report details: https://x.com/poe_platform/status/1922337612720664842