AI power rankings upended: OpenAI, Google rise as Anthropic falls, Poe report finds

The recent use of Poe Openai and Google shows the strengthening of their positions in the main AI categories while losing the Anthropor the floor and showing specialized thinking capabilities as a decisive competitive battlefield.

According to the data issued today, a platform that provides access to more than 100 artificial intelligence model, stake seizures occurred in the important market in all the main artificial intelligence categories between January and May 2025. data, which is extracted from POE subscribers, provides a rare vision in the actual user preferences along with industry standards.

“As a global gateway to more than 100 models of artificial intelligence, POE has a unique vision of the directions of use via the ecosystem,” said Nick Hopper, the lead in the AI’s ecosystem, in an exclusive interview with Venturebeat. “The most surprising things that happen now are fast innovation (3x number of publications from Jan-May 2025 versus the same period in 2024), and an increasing competitive scene, and thinking models are a clear success story in early 2025.”

GPT-4O maintains dominance while new models pick up the market share quickly

In the generation of the basic text, the GPT-4O of OpenAi maintained 35.8 % of the messaging share, while the new GPT-4.1 family in the company has quickly seized 9.4 % of use within weeks of launch. Gueini 2.5 Pro made a Google 2.5 Pro share a message of approximately 5 % after a short period of its presentation.

These gains came largely at the expense of Claude models in humans, which witnessed a 10 % decrease in participation during the reports period. The report indicates that the Claude 3.7 Sonnet has now greatly replaced the former Claude 3.5 Sonite in the user’s preferences, although the latter still maintains a 12 % share of use.

Deepseek, which witnessed a virus growth earlier this year, witnessed its slow momentum as the competitors released logical thinking models at reasonable prices. The Deepsek R1 message from its peak has decreased by 7 % in mid -February to 3 % by the end of April.

Complex problem -solving capabilities become a major discrimination in the artificial intelligence market

Perhaps the most important trend in the report is the massive growth in specialized thinking models, which expanded from about 2 % to 10 % of all text messages sent on POE since the beginning of 2025.

“Thinking models, even in the early days, have shown a great ability to deal with complex tasks with increasing accuracy,” Hopper told Venturebeat. “It is clear that the first adopters find value in this and they are ready to counter the costs in the cost and treatment time to improve results.”

In this high -growth part, Gemini 2.5 Pro soon established itself as a leader, as it acquired about 31 % of the use of the thinking model in only six weeks of launch. It now leads the category, before the specialized models in Claude.

Openai continues to innovate quickly in this field, as it launched multiple thinking models (O1-PRO, O3-MINI, O3-Mini-Hight, O3, O4-MINI) in the first four months of 2025 alone. The report indicates that POE users quickly adopt the latest Openai offers, and the transition from old models such as O1 to newer alternatives like O3.

The report also referred to the emergence of hybrid thinking models, such as Gemini 2.5 Flash Preview and QWEN 3, which can adjust the level of thinking in conversations dynamically. However, these models are currently only about 1 % of the use of the thinking model.

Industry analysts suggest that this shift towards specialized thinking capabilities indicates the mature AI market where raw text generation has become a commodity, forcing service providers to discriminate through high -value capabilities that can require premium pricing.

IMAGEN 3 challenges from Google

The image generation market looks increasingly capable of competition, as the IMAGEN 3 family of Google grows steadily from about 10 % to 30 % of a stake in 2025, which is now competing with the Flux Flux Flux Flux family, which collectively maintains 35 % in late April.

GPT-ISMAAGE-1, which was presented to the application programming interface in late April, achieved 17 % of the use of images generation in just two weeks, reflecting the viral adoption in the Chatgpt app throughout March and early April.

The report indicates that FLUX models have maintained the total number of pluralism in the generation of images on POE, but they witnessed a moderate decrease from about 45 % to 35 % during the reports period.

This tripartite competition between Google, Openai and Black Forest Labs represents a major shift from early 2024, when Midjourney variables and the proliferation variables dominated the area. It has transformed rapid improvement in image quality, adherence to demands, and submitting this category into one of the most patients with artificial intelligence battles.

Institutions have accelerated the generation of images in the past six months, according to supplementary industry data, where marketing departments and creative agencies are increasingly integrated these tools in the progress of production.

Chinese Upstart Kling disrupts the artificial intelligence market

In the generation of the video, the newly released Chinese Laboratory has disrupted the market, as it collectively acquired 30 % use. It is worth noting that Kling-2.0-Master achieved 21 % of all video generation on Poe by the end of April, just three weeks after its release.

Google’s Veo 2 maintained a strong site by about 20 % after its launch in February, while Runway Pioneer Pioneer saw its use of use significantly from about 60 % to 20 % during the reports period.

The speed of the market penetration in KLING highlights the extent of the speed that the competitive scene can turn in the emerging categories of artificial intelligence, as the players who are in charge of their early advantages may not maintain the newcomers quickly and improve.

Video generation remains the most intense artificial intelligence application, as models require a large treatment force to create short clips. This use has kept more limited than generating the text or images, but it is expected that low costs and improvement in quality will increase more accreditation until 2025.

The adoptions in early institutions include advertising agencies, social media content, and educational platforms that have begun to integrate video created by artificial intelligence in their content strategies despite the current technology restrictions.

ElevenLabs dominates the sound of artificial intelligence while new participants target specialized use cases

ElevenLabs continues to lead the sound generation category, as it meets about 80 % of all text requests to speak for subscribers during the reports period. However, the report highlights the competition arising from new arrivals, non -reality speech, Playai, and Orpheus, which provide different audio options, effects and pricing models.

This is the dominance of the market by one player that stands in a flagrant contradiction with the most fragmented competition in other artificial intelligence categories. Industry experts attribute the continuous ElevenLabs leadership to enter the early market, a wide audio library, and high -quality improvements that have maintained a technical advantage over competitors.

New participants find success by targeting the specific market outlets. Unreal Keep gained a traction with podcasts and publishers of Audiobook by presenting specialized vocal representatives and emotional range capabilities. Meanwhile, Cartesia focused on multi -language sounds, and to seize attention from international institutions and educational platforms.

The Audio AI market is expected to grow dramatically until 2025, as the text capabilities are approaching the quality of human being, creating applications in customer service, accessories, and creating content. The relatively low mathematical requirements allow the generation of the video to publish and wider.

The strategic effects of companies that are constantly moving in the scene of Amnesty International

The dynamic nature of the scene of the artificial intelligence model offers the opportunities and challenges facing companies that integrate these technologies.

“It may be difficult to keep up with the latest in artificial intelligence and speed that only gains speed,” Hopper told Venturebeat. “If you are a company already running Amnesty International on a large scale, investing in strong and decorated evaluation pipelines is very important because the best model this month may be the second best next month.”

This volatility in the model preferences confirms the value of platforms such as POE, which provides access to multiple models through one interface, allowing users to compare outputs and adapt to the variable ecosystem of artificial intelligence.

Industry analysts suggest that the increasing importance of thinking capabilities may indicate a shift in how to evaluate companies and spread artificial intelligence models, with an increasing focus on accuracy and reliability of complex tasks instead of just the speed or cost efficiency.

As Frontier Labs continues to launch more capable models at an accelerated pace, companies face difficult decisions about the time to unify specific platforms in exchange for elasticity. Many AI leaders adopt institutions a wallet, using different models of different tasks while maintaining the ability to switch service providers with the development of capabilities.

“This will be an important space for watching, especially among border service providers because it represents the best that artificial intelligence can currently achieve.”

The report indicates that multimedia capabilities have also become increasingly competitive, indicating that the generation of texts, and an essential focus for the development of artificial intelligence, may give way to a more balanced ecosystem where images and videos and generating sound are equally important.

It is possible that companies that successfully move this complex scene are those that maintain the parties to the evaluation that focuses on specific cases of use instead of chasing the latest model versions, while building technical infrastructure at the same time that allows rapid adoption when meaningful improvements appear.

With the continuing models of artificial intelligence in their game of music chairs at the top of the classification, there is one clear thing: in today’s market, the crown rarely remains on the same head for a long time-and companies whose future are betting on the hero of the artificial intelligence may find themselves parallel to Ran Al-Ghad as well.