The following analysis is based on the early 2025 AI ecosystem report published by Poe, a platform developed by Quora that enables users to interact with multiple AI models in one place. This report draws exclusively from usage data on the Poe platform, providing a unique perspective into user behavior and preferences across text, image, and video generation tools. It focuses on how users engage with various models offered on Poe and reflects the growing competition and diversification in the AI ecosystem as of early 2025.
Text Generation Models: OpenAI and Anthropic Lead
In the text generation space, the report reveals a strong duopoly. OpenAI and Anthropic collectively account for around 85% of text-based message traffic on Poe. Notably, Anthropic’s rise in 2024—especially after launching Claude 3.5 Sonnet in June—propelled it close to parity with OpenAI’s flagship models. The growing user base for these expressive and high-performing models underscores their dominance in the expressive text modality.
Meanwhile, open-source models like DeepSeek-R1 and DeepSeek-V3 have gained traction, peaking at a 7% share, surpassing previous favorites such as Llama and Mistral. Conversely, Google's Gemini models, which showed early promise, saw a usage decline after October 2024, suggesting a dynamic and highly competitive environment within Poe’s text model offerings.
Image Generation Models: Market Fragmentation and Innovation
The image generation landscape has become significantly more fragmented. Models such as DALL-E 3 and StableDiffusion that once dominated are now losing ground. Their relative usage dropped by nearly 80%, correlating with an expansion in Poe’s image generation catalog from 3 to over 25 models.
Emerging players have seized this opportunity. BlackForestLabs' FLUX family emerged in mid-2024 and quickly became dominant, holding nearly 40% of the image generation market share on Poe. Similarly, Google's Imagen3—including the speed-optimized Imagen3-Fast variant—grew rapidly after its late 2024 launch, jointly capturing close to 30% usage.
Other providers such as Playground and Ideogram continue to maintain niche user bases, representing about 10% combined usage. This reflects a broader trend: as users seek greater quality, style, and control, they are turning to a wider variety of specialized tools.
Video Generation Models: A Rapidly Emerging Frontier
Video generation, only recently introduced on Poe, has become one of the fastest-growing modalities. Since late 2024, over eight providers have launched video models, bringing a wide range of capabilities.
Runway, launched in September 2024, initially captured the lion’s share and continues to command between 30% and 50% of video generation messages. Google’s Veo-2, one of the latest entries, rapidly climbed the ranks, reaching nearly 40% share within weeks of its release.
A particularly noteworthy development is the rise of Chinese video models, including Kling-Pro-v1.5, Hailuo-AI, HunyuanVideo, and Wan-2.1. These models emphasize performance, lower inference time, and cost-efficiency, collectively securing about 15% of Poe’s video generation traffic.
A Note on Report Scope and Limitations
It’s important to emphasize that this report reflects model usage trends exclusively within the Poe platform. While Poe offers a comprehensive cross-section of leading AI models, its findings do not necessarily represent the entire AI market. Certain models may perform differently or see wider adoption in other platforms, enterprise deployments, or non-English-speaking markets.
Furthermore, model usage on Poe may be influenced by platform-specific integrations, UI/UX design, accessibility, and promotional strategies, potentially skewing comparative data. As such, while the report offers valuable insight into model competition and adoption trends, its scope should be understood within the context of Poe's user base and product architecture.
Keywords
Poe AI report, 2025 AI trends, AI model usage, text generation, image generation, video generation, Claude 3.5, FLUX model, Imagen3, Veo-2, DeepSeek, AI ecosystem analysis
繁體中文摘要
2025 年初 Poe 平台 AI 生態趨勢分析：基於平台使用數據的觀察
本篇文章分析根據 Quora 所開發的 Poe 平台 於 2025 年初發表的 AI 生態系統報告。Poe 是一個集合多個 AI 模型的平台，使用者可於單一介面中與來自 OpenAI、Anthropic、Google、黑森林實驗室等眾多模型互動。此報告以 Poe 平台內部的使用數據為基礎，揭示用戶在文本、圖像和影片生成領域中的行為趨勢及偏好。
在 文本生成模型 方面，OpenAI 和 Anthropic 佔據主導地位，合計佔平台約 85% 使用量。尤其是 Anthropic 於 2024 年中推出 Claude 3.5 Sonnet 之後，使用率快速增長，幾乎與 OpenAI 持平。與此同時，DeepSeek-R1 與 V3 等開源模型也取得了不俗的成績，甚至超越 Llama 與 Mistral 等老牌模型，而 Google 的 Gemini 在 2024 年底後使用率下滑，顯示平台競爭的高度動態性。
圖像生成方面，FLUX 系列 於 2024 年中迅速崛起，並獲得約 40% 的使用率，而 Google 的 Imagen3 也持續成長。儘管 DALL-E 3 和 StableDiffusion 曾經風光一時，但其佔比在短時間內急劇下滑。
在 影片生成 領域，2024 年底開始湧現多個新模型。Runway 率先登場並長期佔據市場份額，而 Google 的 Veo-2 則迅速竄升，短短幾週即奪下近 40% 的使用比例。來自中國的模型如 Kling-Pro-v1.5、HunyuanVideo、萬象 2.1 等則憑藉性能與成本優勢，在 Poe 上獲得約 15% 的影片生成使用量。
值得注意的是，該報告僅基於 Poe 平台內部使用數據。其觀察雖具參考價值，但不代表整個 AI 市場的全面狀況。其他平台或地區的使用趨勢可能截然不同，部分模型表現可能被 Poe 的用戶結構、介面設計與行銷策略所影響。因此，該報告應以「平台內部觀察」的角度理解其趨勢結論。
Reference
Poe Blog (2025). Early 2025 AI Ecosystem: Growth in Text, Image, and Video Models. Retrieved from https://poe.com/blog/report-early-2025-ai-ecosystem-trends
0 Comments