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

PoeMost recent Use report joyous Open and Google To strengthen positions in KEY AI categories Anthropical The ability of the ground and specialized justification appears as a limited competition.

According to Information released by PoE todayA platform offers access to more than 100 AI models, a significant market share shifts in all major AI categories between January 2025 and May. The information drawn from Poe subscribers provides unique visibility to actual user choices beyond industrial criteria.

“Like a universal gateway to 100+ AI models, Poe has a unique view of the trends in the ecosystem,” said Nick Huber, PoE’s EC Ecosystem management, in an exclusive interview with Ventürebeat. “The most surprising events that are currently a fast innovation (in January-May 2025 in 2025 in 2025), increasingly various competitive landscape and justification models are a clear success story in early 2025.”

GPT-4O, new models are dominated when capturing market share quickly

In the main text generation, Openai GPT-4O The company will be newer, maintained the command position with 35.8% of the message share GPT-4.1 Family Quickly applied 9.4% of the use during the start weeks. Google’s”s Gemini 2.5 Pro Similarly, a short time with the application was achieved about 5% of the message.

These gains have come mainly at an anthropic account Cleaning modelsThose who see 10% definitely in the reporting period. Notes in the report Claude 3.7 Sonnet now substituted the previous significantly Claude 3.5 Sonnet Although in the user option, the last still keeps the share of 12% of the use of 12%.

DepthEarlier this year, the virus lived in the growth and slowed its speed because of their own favorable, venture models. DeepSeek R1By the end of April, the message share of 7% to 3% in mid-February decreased.

Complex problem solving opportunities are the main distinguishes in the AI ​​market

Perhaps the most important trend in the report has a sharp increase in special thinking models, from about 2% to 10% of all text messages sent to POE since 2025.

“Days have shown that so far, even substantiated models, showed a remarkable abilities to manage complex tasks with increasing accuracy,” said Huber Ventureat. “Early adopts find clear information in this and are ready to take the cost and processing time to get better results.”

In this high growth segment, Gemini 2.5 Pro He quickly identified himself as a leader, seized about 31% of the model use inside only six weeks. Now leads the category of clod-based models.

Open It continues to innovate in this space, leaves multiple models of thinking (O1-PRO, O3-Mini, O3-mini-high, O3and o4-mini) alone in the first four months of 2025. The report shows that POE users pass through the old models to the newer alternatives, and passed the latest proposals to OPED and quickly adopted OPP.

The report was noted as in the case of hybrid justification models Gemini 2.5 Flash review and Qwen 3Negotiations can be dynamically adjusted. However, these models are currently only 1% of the model use.

Industrial analysts have signed this change in the transformation of this change in special thinking, the re-establish of raw text generation, and to distinguish the premiums of providers with higher value opportunities.

Google Image 3 Difficulty Wisual AI Arena set up players

Description Generation Market Google seems to be increasingly competitive Figure 3 Family from about 10% to 30% in 2025, now the rival category leader Black forest laboratories Flux family About 35% shared models by the end of April.

Openai GPT-Image-1In the late April, the image was presented to API, only 17% of the image generation in two weeks, quickly obtained viral adoption Chatgpt application March and early April.

The report indicates that Flux models In POE, the imagery has maintained a total majority shared in the generation, but in the reporting period, about 35% of about 45% to 35% to 35%.

These three-way competition Google, Openand Black forest laboratories Marks an important turn since the beginning of 2024 Midjourney and Fixed spread Options dominated in space. Fast improvement in the quality of description, urgent and adherence to the showing speed turned this category to one of the most violent controversial AI battlefields.

The adoption of the enterprise’s generation, in the past six months, according to additional industry data, the production of marketing departments and creative facilities, the production of these tools in the past six months has significantly accelerated.

China UpStart Kling Video AI market breaks the early lead of a difficult runway

Video generation, Chinese laboratory Kuuishou’s Newly released Kling family The models quickly broke the market quickly by collecting about 30% of the use of about 30%. Most importantly, Kling-2.0-ASTER, until the end of April, received 21% of the entire video generation in PoE only three weeks after release.

Google’s”s I see 2 After the start of February, he took a strong position with a share of about 20%, pioneering Runway During the reporting period, he saw that his use was reduced by about 60% to 20%.

Speed Kling’s The market reputation emphasizes how the competitive views can change rapidly in the formation of the AI ​​categories, and the newcomers of the built-in players can not quickly intercourse and develop.

Video generation remains the most report-intensive consumer-face AI application that requires significant processing power to create the most short clips. This limited the text or offspring generation, but expected to improve the expenses and quality of rapidly and until 2025 is expected.

Early enterprise adopts, advertising agencies, social media content creators and the current restrictions of technology, include advertising agencies and educational platforms that start to access content content in the AI.

Elevenlabs is dominated by a voice AI when targeting new applicants

Onelabs During the reporting period, all subscribers continue to lead a voting category, fulfilling about 80% of the text’s requirements. However, the report emphasizes the competition of newcomers CARTY, Speech, Playaiand OrfeyDistinguished sound options, effects and price models.

The advantage of this market by a player stops more tougher than more fragmented competition in other AI categories. Industrial experts continue to improve the sequential quality of Elevenlabs’ early market access, extensive voice library and rivals.

New applicants find good luck by targeting special market marks. Speech Podcast manufacturers and audiobook publishers by offering special sound actors and emotional range capabilities and audiobook publishers. Meanwhile CARTY Showing interest from global enterprises and educational platforms, he payed attention to multilingual voices with original accents.

The Audio AI market is projected to grow significantly until 2025, until 2025, and the text approached the quality of the text and to find applications in customer service, accessibility solutions and content work. Compared to the generation of video allows relatively low computing requirements for more placement and experience.

Strategic results for businesses that are constantly changing the changing AI view

The dynamic nature of the AI ​​model landscape provides both opportunities and difficulties for enterprises that integrate these technologies.

“It can be difficult to continue the latest in the EU and PACE only gains speed,” said Huber Ventureat. “If you are already working on a large-scale AI, strong, provider-agnostic evaluation pipelines are critical, because the best model this month can be the second best month.”

This volatility in the model benefits is the value of platforms such as POE, which includes several models via an interface that allows users to compare access to access and change the changing AI ecosystem.

Industrial analysts can signal a turn of increasing importance, not only increased importance, not only speed or expense efficiency, not only speed or expense efficiency, but increased attention and reliability.

As border laboratories continue to leave more skilled models in a more skilled pace, enterprises face difficult decisions on specific platforms protecting comfort. Many enterprises accept a portfolio approach to various tasks to protect the capacity to develop strengthen, providers of the AI ​​leaders.

“This is especially among the border providers, especially between border providers, because it is currently the best of what it can represent,” It is noted about Huber models.

Despite the fact that multimedia opportunities are increasingly competitive in the report, a more balanced ecosystem, which is the initial focus of the AI ​​development, can be made to the same extent of the image, video and audio generation.

The successful managers that successfully manage this complex picture are likely to build the latest model releases, but also to build technical infrastructure, and protecting the evaluation framework for the construction of the technical infrastructure, which allows meaningful developments.

Following the rating of AI models, the AI ​​models are watching the rating: in today’s market, the crown rarely, and the companies that bet the future with yesterday’s AI champion can match them tomorrow.