Text Generation Trends
A Two-Way Race
The study shows that among Poe subscribers, Anthropic models are quickly becoming as popular as OpenAI, especially after the release of Claude 3.5 Sonnet. The usage of text models from both providers is now almost evenly split.
Rapid Adoption of New Releases
Poe users often switch to the latest models, even if loyal to a specific brand. For example, people rapidly move from OpenAI’s GPT-4 to GPT-4o or from Claude 3 to Claude 3.5.
Emerging Players
DeepSeek’s R1 and V3 have captured about 7% of the messages on Poe. Google’s Gemini family has seen a slight decline in use among Poe subscribers but remains a key player.
Image Generation Trends
Market Share of Early Movers
DALL-E-3 and StableDiffusion were once leaders in image generation, but their shares have dropped by about 80%. This decline occurred as the number of image generation models increased from three to around 25.
Leading Models
The FLUX family from BlackForestLabs is now the leading image model, holding a nearly 40% share, while Google’s Imagen3 family has about a 30% share.
Smaller Models
Smaller image providers like Playground and Ideogram update their services frequently, which helps them maintain a loyal user base. However, they only account for about 10% of Poe’s image generation usage.
Video Generation Trends
An Emerging Industry
Video generation was almost nonexistent on Poe until late 2024, but it has quickly grown in popularity. Now, at least eight providers offer this ability.
Runway: Most Used Model
Runway’s single video model handles 30–50% of video generation requests. Although its usage is lower than it used to be, many people still choose this brand.
New Player: Veo-2
Since launching on Poe, Google’s Veo-2 has gained about 40% of the market, showing how quickly customer preferences can change. Other new models, such as Kling-Pro v1.5, Hailuo-AI, HunyuanVideo, and Wan-2.1, have captured around 15% of the market.
Key Takeaway & Looking Ahead
The data shows a clear pattern of newer models replacing older ones in user preference. If you want the best performance, use the latest version rather than stick with familiar but outdated models.
Whether these usage patterns will hold steady or continue to shift remains to be seen. At some point, cost will be a barrier to adoption, as new models tend to get more expensive with every release.
In future reports, Poe plans to share insights on how different models fit various tasks and price points.
FAQs
Q: Which models are most popular for text generation?
A: The study shows a two-way race between Anthropic and OpenAI models, with both providers having an almost even split in usage.
Q: How quickly do users adopt new models?
A: Poe users tend to rapidly move to the latest models, even if loyal to a specific brand.
Q: Who are the emerging players in image generation?
A: DeepSeek’s R1 and V3, and Google’s Gemini family, are among the emerging players in image generation.
Q: What is the current state of video generation?
A: Video generation is an emerging industry, with at least eight providers offering this ability, and Runway’s single video model holding a significant market share.
