https://chatgpt.com/
https://www.deepseek.com/
https://character.ai/
www.perplexity.ai/
https://janitorai.com/
www.anthropic.com/claude
https://quillbot.com/
https://suno.com/home
https://spicychat.ai/
https://www.doubao.com/chat/
https://www.moonshot.cn/
https://hailuoai.video/
https://huggingface.co/
https://poe.com/
https://adot.ai/multillm
https://edenai.world/
https://www.polybuzz.ai/
http://seaart.ai/
https://getliner.com/zh
https://klingai.com/
https://civitai.com/
https://elevenlabs.io/
https://openai.com/sora/
https://crushon.ai/
https://www.blackbox.ai/
http://deepai.org/
https://gamma.app/
https://leonardo.ai/
http://cutout.pro/
https://brainly.com/
https://www.photoroom.com/
https://moescape.ai/
https://www.midjourney.com/
https://candy.ai/
https://zeemo.ai/cn/
https://www.veed.io/
https://invideo.io/
https://www.pixelcut.ai/
https://www.talkie-ai.com/
http://pixai.art/
https://monica.im/
https://www.cursor.com/cn
https://ideogram.ai/
https://chub.ai/
https://clipchamp.com/en/
https://www.meta.ai/
https://studyx.ai/
https://bolt.new/
https://picwish.com/
https://www.joyland.ai/
/
This list focuses on AI native apps and strictly excludes traditional tools that only overlay generative AI features (e.g. Canva, Notion, etc.).
Compared to previous editions, we further adjusted non-AI native photo editing products such as Pixlr, Fotor, PicsArt, etc. out of the list. We've also added a new “waitlist”: 10 companies that unfortunately didn't make the cut - 5 web products and 5 mobile apps - and we're curious to see which (if any!) products will be in the top 10 next time around. We're curious to see which (if any!) products will make the top 100 next time around.
ChatGPT came out of nowhere with a research preview version in November 2022, capturing 100 million users in just two months, setting a record for the fastest-growing consumer-grade app. But it didn't last long, and between March 2023 and April 2024, its global monthly activity almost came to a standstill. The turning point came in 2024: it took ChatGPT nine months to go from 100 million weekly users in November 2023 to 200 million in August 2024, and less than six months to go from 200 million to 400 million weekly users in mid-February 2025. Typically, the larger the user base the harder it is to grow, but ChatGPT bucked the trend and achieved an accelerated breakthrough in the scale-up phase.
Growth of the GPT web side
The initial growth stagnation was due to the “taste effect”: although users found ChatGPT novel, it was difficult for them to find daily use scenarios. However, as OpenAI continues to introduce more advanced models and features, ChatGPT's usage rate rises, attracting new users and increasing the activity of old users.
The growth curve highly overlaps with product milestones: the April-May 2024 spike stems from the release of GPT-4o, whose multimodal capabilities allow users to have real-time conversations, display pictures, and get intelligent feedback - from taking photos of math problems to AI brainstorming, the experience is comprehensively upgraded.
The July-August 2024 jump is attributed to Advanced Voice Mode going live, with AI conversation fluency approaching that of real people. And the September-October growth spike? It was the debut of the o1 model series, which took its reasoning and problem-solving capabilities to the next level.
GPT Mobile Growth
Since its launch in May 2023, ChatGPT mobile monthly active users have always maintained a steady growth of 5%-15%, with a large number of web-based users shifting to the mobile app. According to Sensor Tower's estimation, out of ChatGPT's 400 million weekly active users, the mobile share has reached 175 million.
DeepSeek is a double whammy: even though its public beta only went live on January 20, it took just 10 days to rank second on the global monthly traffic list for AI products, a stunning record.The product, developed by a Chinese hedge fund, had its January 2025 users mainly from China (21% share), which is exactly where ChatGPT is off-limits. Other big users include the US (9%) and India (8%). However, DeepSeek has also encountered bans or restrictions on government equipment in South Korea, Australia, Taiwan, and several U.S. states.Compared to other general-purpose big language models, DeepSeek's start has been phenomenal: it captured 1 million users in 14 days (though slightly behind ChatGPT's 5-day record), but surpassed the 10 million mark in just 20 days, significantly ahead of ChatGPT's 40-day milestone.
DeepSeek sees significant growth in monthly unique visitors
Launched on January 25th, DeepSeek was ranked 14th on the monthly activity list in just five days. By February, its ranking had jumped to the second place, and the mobile user scale reached 15% of ChatGPT.Sensor Tower data shows that DeepSeek is not a “shell download”: in terms of average weekly usage and duration, its user activity is slightly higher than that of Perplexity and Claude, but there is still a significant gap compared to ChatGPT.
DeepSeek Mobile User Activity AnalysisComparable to competitors, not as good as ChatGPT
DeepSeek's rapid rise in popularity in the tech and research worlds has been fueled by its impressive performance in reasoning benchmarks and its claim to have cost only $5.6 million to train - a fraction of its competitors' costs.The training cost statement attracted widespread media attention, and mainstream opinion even compared it to the “Sputnik moment” in the field of AI. Google Trends shows that on January 27th, DeepSeek's global search popularity was the same as that of ChatGPT, and even surpassed that of ChatGPT in the U.S. The company's success in the field of AI has been recognized by the media and research community!!
AI video has been teetering on the edge of being truly usable for the past 18 months, but the last six months have finally seen major breakthroughs in quality and controllability.
For the first time, three companies - Hailuo AI (#12), Kling AI (#17), and Sora (#23) - made the site rankings, alongside InVideo (#37), while Runway and Krea (which supports cross-model generation of video) made the waitlist.
Hailuo AI (developer of the MiniMax series of models) and Kling are both Chinese video models - Kling was released in June 2024, with Conch following in September. As of January 2025, both have surpassed Sora in terms of monthly visits, and while Sora put out a preview version as early as February 2024, it waited until December for its official release.Video modeling providers are moving towards differentiation: whether in terms of output style or user interface, each is exploring unique positioning and offering more options for creators. sora is known as a versatile, all-in-one video generator, conch specializes in accurately restoring cues, and Kling offers additional features such as camera motion control and lip sync.
Comparison of three model generation(Cue word: Corgi running in a field, photo-realistic)
AI video editing remains the top use case for consumer-grade AI, simplifying time-consuming tasks such as smart editing and caption generation into “one-click” solutions.
Veed (#36) and Clipchamp (#45) made the website rankings, while mobile was dominated by video-photo hybrid editing apps represented by B612 (#12), VivaCut (#15) and Filmora (#19). In terms of revenue, Splice, Captions and Videoleap were the brightest performers. More changes are expected to come to this space in the coming months, with models like Google's Veo 2 (not yet publicly available) expected to raise the industry standard based on early beta feedback. However, with a pricing of $0.50 per second, Veo 2 is likely to be skewed more towards commercial applications than the mass consumer market.
It's a time for builders, and AI tools are rising to meet the demand. In the past six months, two types of products have emerged, targeting different user groups:
(1) Intelligent IDEs (integrated development environments) for developers;
(2) text-to-web application platforms (Andrej Karpathy calls them “vibecoders”) that can be used by both technical and non-technical users.
Intelligent IDEs such as Cursor, which recently announced that it has served hundreds of thousands of developers, play the role of developer co-pilot, providing error checking, auto-completion, and full code generation. In contrast, text-to-web application products do not require users to have a technical background. Simply type in a text prompt and you get a usable Web site or Web application, as if by magic. The rise of these products is due to multiple factors: models that can finally generate executable code, the growing maturity of Web frameworks, and platforms that can leverage libraries and SDKs from companies like Resend, Clerk, and Supabase.
Similar to smart IDEs, the leaders in this space are growing rapidly. bolt made its debut on the list at #48, with Lovable making the waitlist. Bolt reported $20 million in annualized revenue and 2 million registered users in its first two months, while Lovable reached $17 million in annualized revenue in its first three months. The top two smart IDEs are now ahead of the top two no-code platforms in terms of monthly unique visitors (not visits) to their websites, but both are growing at comparable rates. We look forward to seeing how this landscape evolves. Smart IDEs require users to have a technical background, but every developer can theoretically get near daily use out of them. In contrast, no-code platforms are aimed at 99% of non-coding users, but require users to have a clear idea of what they want to build.
Intelligent IDE vs. Text-to-Web Application Builder
There is some overlap in the user base of these two types of products. Technical users may utilize the no-code platform for prototyping or generating downloadable and refined code.Similarweb data shows that 23% of unique visitors to Bolt in January 2025 also visited Cursor.
Overlap of Bolt and Cursor usersAverage Unique Visitors Statistics for January 2025
The mobile app rankings in this article are based on the number of monthly active users (MAUs) worldwide, as tallied by Sensor Tower. However, the most popular apps aren't always the highest earners. To make this list, apps need to have more than 8 million MAUs, but many apps with far fewer users are achieving higher conversion rates and user revenues.ChatGPT is way ahead in both metrics. However, in our analysis of the top 50 MAUs with the highest monthly revenue (from mobile subscriptions) apps, only 40% overlap.
Of the 80 apps that ranked among the top 50 in terms of revenue or usage, many of the lower-usage apps proved to be the most effective cash generators - at least through mobile subscriptions, since ad revenue was not accounted for.
Top AI Consumer App Usage vs. Revenue AnalysisData source: mobile subscriptions
Some product categories are represented in both lists, but by different companies. Photo and video editors make up 20% of the revenue list and 24% of the MAU list. However, the top three most used video editors - VivaCut (#15), Filmora (#19), and Beat.ly (#36) - and the top three highest-earning video editors, Splice, Captions and Videoleap do not overlap. The apps with the most mass appeal don't always have the advanced semi-professional or even enterprise-level features that more specialized users are willing to pay for. Some product categories that only appear on the revenue list and are missing from the usage list include plant identification (PictureThis, PlantID), nutrition (Cal AI, Fastic), language learning (Speak, Learna, Loora), music (Moises, Suno) and dictation (Otter, PLAUD). These apps target specific goals or skills and, while less generic, are more likely to attract users willing to pay for their specialized value.
Interestingly, ChatGPT “copycat” apps make up 12% of both the mobile usage and revenue charts. We first reported on these apps in our mobile rankings, noting that by mimicking ChatGPT's name and logo, they implied that they offered the same premium model at a lower price. They often tweak their app listings to circumvent app store enforcement and copyright claims.