Poe has just unveiled a complete study on the use of the 100 AI models it provides its users. The disparities in use between the different methods are striking.
Who really leads to generative AI in 2025? According to data for use of the latest POE report, the generative IA platform of Quora, the landscape is changing. If Openai and Anthropic share 85% of the text market (from 01/24 to 02/25), other actors like BlackForestlabs with flow in the image and Google with VEO-2 in video jostle the established order.
In the text, Openai and Anthropic dominate without surprise
From January 2024 to February 2025, Openai and Anthropic dominate the market for text models, representing together around 85% of the textual prompts addressed to Poe. Notable fact, since the launch of Claude 3.5 Sonnet in June 2024, the use of anthropic and Openai models has become almost equivalent.
The emergence of new actors nevertheless upsets this balance. The Chinese Deepseek, who had no presence in December 2024, managed to gain 7% of market share at its peak, a level significantly higher than the previous families of open-source models like Llama and Mistral. Meanwhile, Google, after initial growth with its Gemini family, has a decline, while the long train of the “other” official models (like those of XAI) continues to expand, testifying to an increasing diversity in the ECA ecosystem. The supremacy of the Giants owners of the AI does not now seem acquired.
Trendy reversal in image generation
The historic pioneers of the image generation have had their time. Since January 2024 and until February 2025, Dall-E-3 and the different versions of stablediffusion (Midjourney is not available on Poe) have seen their share of use drastically drop by almost 80% since February 2024. A spectacular decline which coincides with the explosion of the number of models, going from only 3 to around 25 in a few months. BlackForestlabs achieved a remarkable coup with his family of flow models published last spring. In just a few months, Flux has become the dominant model of image generation by capturing almost 40% of prompt.
Google is not to be outdone with Imagen 3, launched at the end of 2024, which quickly won a significant market share, with almost 30% of uses. Finally, specialized providers such as Playground and Ideogram remain relevant, but collectively represent only 10% of market share. Difficult to compete against Google and BlackForestlabs.
Runway and Google in ambush in the video
The generation of video by artificial intelligence remains a still embryonic market which experienced dazzling growth at the end of 2024 and early 2025, recalls Poe, which analyzes the period from November 2024 to February 2025. Really structured since the end of 2024, the sector has quickly diversified, going from a few providers to more than eight players offering various solutions on Poe. Runway, which launched its model in September 2024, initially dominated the market, capturing between 30% and 50% of video generation prompts, despite a slight erosion on its market share over time and the supply of a single model in its API.
In 2025, Google made a shattering entry with VEO-2, which managed to capture almost 40% of prompt video generation only a few weeks after its launch on Poe. Note that Sora d’Openai not having a API, it is not available on Poe. At the same time, Chinese models such as Kling-Pro-V1.5, Hailuo-Ai, Hunyuanvideo and Wan-2.1 continue to repel borders in terms of capacities, inference and cost time. Chinese models represent around 15% of prompt video generation. They gradually gain market share, without so as to compete with heavy goods vehicles in the sector.
Very concretely, the owner technological giants (Openai, Anthropic and Google mainly) retain their dominant position. However, the lines are moving: start-ups are gradually gaining ground, while the Chinese ecosystem is distinguished by its rapid innovation and its ability to offer models at competitive costs.
As a reminder, the report data only reflects the habits of POE users and do not give an exhaustive panorama of the market. Some major models, such as Sora d’Openai or Midjourney, which do not offer an API, are naturally absent from the ranking. Despite these limits, the analysis offers an interesting overview of major trends in the sector.
