Original · Unique Research · 2026-08-12
Editor's note: The first-person analysis and all data figures belong to the original Chinese author. This English rendition retains the four signal sections, closing analysis, and methodology notes. All named companies, products, and figures are preserved. Traffic data, model specifications, and earnings figures are as reported by Unique Research, not independently verified for this rendition.
AI Data Insights
The Truth About July's AI Data
"Model capability determines the ceiling, but entry point, compute, and cost determine who keeps the traffic."
Start with three numbers:
+110.4%: A product already at 15M MAU doubled its monthly active users in July.
-56.9%: A product backed by the world's largest software company halved its MAU in a month.
+26157.8%: The global growth chart's #1 this month — but it's not really growth, it's "moving house."
These three numbers belong respectively to Kimi, Microsoft Copilot, and Baidu's Wenxin.
These days the AI news cycle is dense: Google's AI leadership change, Manus separating from Meta to restore independence, Musk's $60B acquisition of Cursor followed by Grok Bot. But July's traffic data told the story earlier. Below is Unique Research's statistical record of global and domestic AI Web data for July 2026 — we break it apart and examine the signals behind the numbers.
Signal One: Giants Lose Speed Collectively, Google's Trouble Is in the Data
Start with the global top-tier traffic landscape:
| Product | July visits | MoM | July MAU | MoM |
|---|---|---|---|---|
| ChatGPT | 5.45B | +1.27% | 469M | -9.07% |
| Gemini | 2.687B | -6.03% | 302M | -6.67% |
| Claude | 969M | +2.38% | 108M | +1.27% |
| Grok | 210M | -0.90% | 30.92M | -0.57% |
| Perplexity | 104M | -10.38% | 18.71M | -6.03% |
ChatGPT remains in first place by a wide margin — its visits are more than twice the #2, Gemini. But note a divergence: while visits ticked up 1.27%, MAU dropped 9.07% — fewer people came, but those who came came more often.
Google's apps have another headache.
Gemini visits -6.03% MoM, MAU -6.67%, while June's figures were -1.51% and -3.2% respectively — the declines are widening and have been in a downward channel for months. NotebookLM fared even worse: visits -26.77%, MAU -21.98%. Google AI Studio stayed flat.
It's not surprising after months of declines: the flagship model Gemini 3.5 Pro was delayed repeatedly, and the Flash series had to carry the show — netizens nicknamed it "North American Soybean Bun." Model delivery delays and talent departures culminated in the August 5 leadership shakeup: Hassabis stepped down as DeepMind CEO to become chairman, 27-year veteran Jeff Dean left to start a company, and Alphabet's stock dropped about 4% intraday. The successor, Kavukcuoglu, comes from a DeepMind CTO background, leaning toward engineering execution — the signal is clear: Google doesn't want the next AlphaFold, it wants a Gemini that delivers on time.
Microsoft's Copilot wasn't much better: July visits -31.35%, MAU halved (-56.92%), from 38.19M to 16.45M. In 2026 when everyone is competing on Agents, Copilot's Web side is being rapidly marginalized.
Claude is the only product in the top tier with positive double-metrics momentum. The reputation built in coding and Agent scenarios is steadily converting into traffic.
Signal Two: Kimi's July Was the Fiercest Month for Domestic AI in Recent Years
The biggest variable in July's data is Kimi.
kimi.com visits 60.1M, +52.19% MoM; MAU 15.37M, +110.36% MoM — literally doubled. Among the global top 20 MAU growth chart, it's the only product over 10M in size that still doubled.
On July 16, Moonshot AI released Kimi K3 — 2.8 trillion parameters, 1M token context, the world's largest open-source model. Within hours of release, it topped the Arena coding leaderboard — the first time a Chinese large model took that position. Musk commented "Impressive" under the related evaluation, and NVIDIA's market cap briefly lost $111.1 billion in a single day.
What happened next says even more about demand intensity: 48 hours after release, Kimi suspended new user subscriptions due to compute shortage, prioritizing existing users. An AI product "rate-limiting" because too many people want to use it — the first time in the industry. At the end of July, Bloomberg also reported that Moonshot AI had reached a compute agreement with Alibaba to use 20,000 NVIDIA chips (Alibaba denied the "H200" specification but did not deny the chip supply itself). The capital side is also surging: Kimi's ARR surpassed $300 million in mid-June, and the new round has a pre-money valuation of $31.5 billion — up more than 6x from the $4.3 billion in December 2025.
Third-party platforms are corroborating. On OpenRouter, global weekly token calls hit consecutive three-week highs in July, from 46.7 trillion to 62.3 trillion — K3 was one of the biggest drivers in the third week. Chinese model weekly calls have exceeded US models for 12+ consecutive weeks, reaching a 63.5% share in the trailing 28 days by end of July — of every 10 model calls globally, over 6 use a Chinese model. The explosion of domestic models has also spilled into the infrastructure layer: SiliconFlow, doing inference services, saw July visits +109.12%.
In the same July, two other domestic top-tier companies' situations serve as a control group:
Qwen also released a new model (Qwen3.8-Max preview on July 19, 2.4T parameters), but it was prioritized for developer and office channels like Qoder and Token Plan, not the C-end chat box. The result: domestic Qwen visits -18.81%, overseas Qwen -4.01%, while Qwen Cloud targeting agent scenarios surged +901.97%, hitting #2 on the global growth chart. Same release, C-end down, B-end up — Alibaba's重心 this month was clearly not on the chat box.
DeepSeek's new model was "delayed," no relay, July visits -6.02%, MAU -11.26%, second consecutive month of double decline. Its V4 Flash official version only appeared on OpenRouter in early August, and single-week calls immediately surged to 8.83 trillion tokens (+570%) to #1. There is no more direct experiment on how much "releasing a new model" matters for traffic.
Tencent released the Hunyuan Hy3 official version on July 6, using MoE architecture with 295B total parameters, 21B active parameters, supporting 256K context, open-sourced under Apache 2.0. This model's positioning is interesting — it shows intelligence levels significantly stronger than same-size models, even comparable to flagship models with 2-5x parameter scale, while API pricing is compressed to 1 RMB input / 4 RMB output per million tokens. After release, Tencent Cloud TokenHub platform daily token calls grew 5x within two months, exceeding 25 trillion. Tencent also previewed Hy4's upcoming launch, further boosting multimodal capabilities. If K3 is the "world's largest open-source" banner, Hy3 takes the pragmatic "high cost-performance + open-source" route — the two formed a "dual-track" domestic model landscape in July.
Incidentally, Kimi's run also changed the domestic rankings: 60.1M monthly visits already exceeded Baidu AI Search (29.96M) and Wenxin's new entry (29.34M), second only to DeepSeek, Doubao, and Quark; the gap in MAU with Quark narrowed to under 6M. At this slope, the domestic first tier will soon be re-ranked.
Signal Three: From "Chatting" to "Working" — the AI Work Track Starts Scaling
The third signal is hidden in the mid-tier. A batch of "AI getting work done" products are ramping up, and they're all backed by giants:
WorkBuddy-CN: Entered the chart at 6M visits and 1.68M MAU, parachuting directly to #14 in the domestic visit chart. Elevator screens are also full of its ads — will it be the next national-grade AI app?
Tencent's 2026 Q2 earnings provided a more authoritative official perspective on WorkBuddy. The earnings disclosed that WorkBuddy PC-side June visits already exceeded 20M, ranking first domestically among similar products, exceeding the combined visits of the #2 and #3. Citi Research's data shows MAU exceeding 20M and DAU exceeding 13M. Tencent has internally positioned it as the third phenomenon-level product after QQ and WeChat, first to achieve full coverage across iOS, Windows, Android, and HarmonyOS. The technical backbone behind this is Hunyuan Hy3 — in real business scenarios like WorkBuddy, CodeBuddy, and Marvis, Agent task success rate improved from 72% to 90%. Ma Huateng directly commented in the earnings that WorkBuddy and CodeBuddy achieved breakthrough user growth and are at China's leading level in their respective fields. Notably, management's attitude: the enterprise version started raising prices in May, but Tencent explicitly said "no commercialization KPI has been set for the team; it remains in the investment phase." This forms a subtle contrast with Alibaba's integration of Qwen Office and ByteDance's splitting of Feishu — Tencent chose to saturate single-point investment rather than multi-point probing. CodeBuddy's data is equally impressive: internal Tencent validation reduced engineer coding time by 40%.
Xiao Huanxiong (SenseTime): visits 9.23M (+158.51%), MAU 4.64M (+152.45%) — the fastest-growing near-10M-level product domestically.
MuleRun: visits +147.91%, MAU +138.08% — this "going-overseas" agent is actually an Alibaba product; on August 3 it was combined with QoderWork and Wukong into "Qwen Office" for public beta, led by new DingTalk CEO Chen Yusen.
codebuddy-CN: MAU 3.3M, +45.91% MoM.
Zhipu's ZCode newly entered the chart (1.82M), Zhipu Qingyan +27.63%; StepFun AI +42.35%, iFlytek Spark +41.53%, Metaso AI Search +23.04%.
Looking further out: Claude Cowork went on cloud a month ago, Musk released Grok Bot this week, and OpenAI is giving Codex the ability to control other applications. Domestically, ByteDance split the Feishu team in two at end of July, merging into Doubao and Volcano Engine respectively. All vendors are ending "horse racing" and consolidating Agent capabilities into office-facing flagship products.
CICC summarized this trend as from "Vibe Coding" to "Vibe Working": the user base expands from 100 million programmers to 600-700 million office workers. July's mid-tier data is the first traffic-level footnote to this statement.
Signal Four: The Quiet Growth Club — Only 6 Products Above 5M MAU Still Growing Double Digits
Most top-tier products are declining. Were there products in July that are "big enough and still growing fast"? We screened the global MAU chart: above 5M MAU, double-digit MoM growth, excluding Wenxin's caliber distortion — only 6:
| Product | July MAU | MoM |
|---|---|---|
| Kimi | 15.37M | +110.4% |
| candy.ai | 32.86M | +14.2% |
| Higgsfield | 8.29M | +15.6% |
| GPTGirlfriend | 6.49M | +32.6% |
| Meshy | 5.52M | +13.0% |
| OpenRouter | 5.46M | +14.5% |
Emotional companionship took two seats, and it was the steadiest track in July. candy.ai MAU 32.86M, higher than DeepSeek; JanitorAI visits +12.86%, Emochi +31.55%, ourdream.ai +12.15%, CrushOn +15% — in a July where tools were mixed, emotional companionsity was almost universally green. The first stable money AI makes is probably still "emotion."
Video is分化: the strong ones are Higgsfield and HeyGen. Higgsfield visits 31.32M (+16.8%), MAU 8.29M (+15.6%) — the largest and fastest-growing in video; HeyGen MAU +16.8%. Kuaishou Kling visits only +1.7%, and ByteDance's is weaker: Dreamina visits -17.88%, MAU -20.29%, Jimeng -5.34%. Overseas tools are up, while domestic giants' video products are down. Notably, ChatCut surged to 170K active users in July, +292.87% MoM — the explosion comes from integrating the Codex-cli plugin to enable Agent natural-language control of editing. It lets Codex directly operate an editable timeline, one-click completing gap removal, rough cut, subtitles, soundtrack, and B-roll configuration; creators only give text instructions and then fine-tune the timeline. Video creation is entering a new trend: repetitive editing drudgery handled by AI agents, creators freed from software operations to focus on content planning, narrative pacing, and aesthetic control. Agent-driven natural-language post-production will become the mainstream workflow for short-video creators and knowledge creators.
Two "shovel sellers" are also up: 3D modeling Meshy MAU +13.0% (same-track Rodin MAU +107.89%); model aggregation platform OpenRouter visits +11.26%, MAU +14.5% — developers are voting for Chinese models with real money in call volume.
About Wenxin's "26,158% Surge," Let's Be Honest
July's global traffic growth chart and MAU growth chart both top-ranked Baidu's Wenxin: visits +26,157.77% MoM, MAU +21,740.18% MoM.
This number needs to be put in giant quotation marks.
What actually happened is Baidu completed entry integration in July: the two old domains chat.baidu.com and yiyan.baidu.com automatically redirect to wenxin.baidu.com. Breaking it down:
chat.baidu.com: visits -32.69%, MAU -24.78%
yiyan.baidu.com: visits -53.04%, MAU -22.34%
wenxin.baidu.com: visits grew to 29.34M
Three domains without cross-domain dedup, directly added: visits from 46.94M in June to 60.39M in July, +28.65% MoM. The old entry lost about 15.78M visits; the new entry received about 29.23M — migration basically holds, but the redirect chain may cause the same visit to be double-counted. This combined growth is under "no cross-domain dedup caliber" and cannot be equated with real growth.
"This isn't Wenxin gaining 261x users in a month; it's Baidu hanging traffic scattered across three doorplates onto one new sign."
We're calling this out separately to emphasize a principle: when looking at charts, don't just look at growth rate — caliber changes matter more than the numbers themselves.
August Will Bring Bigger Variables
July's data is done, and August's cards are already being shuffled.
First, Manus. On August 11, the Agent star acquired by Meta seven and a half months ago announced it was restoring independence; Tencent and other original shareholders bought back the shares for $2 billion. But netizens' reaction was rather lukewarm. Someone commented: "Nobody cares anymore, a remnant of the old era." Another was more direct: "Not as good as Codex Pro. No model of its own, can only parasitize on others' models to do Agents — sooner or later it'll be eaten dry by model vendors."
Rough words but sound logic — data is on the netizens' side: in the seven and a half months after acquisition, Manus visits declined continuously (June -17.19%, July -15.72%, MAU -21.99%). After the honeymoon, there was a long giveback. Restoring independence only gets the right to re-enter the board; the sharp question remains: after model vendors go directly into Agents, how much value is left in the middle layer?
Three more cards:
Grok Bot on duty: cloud computer + multi-bot collaboration + follower learning, directly benchmarking Claude Cowork. Grok's Web side stayed flat in July (-0.90%); whether Grok Bot can pull it back to growth, August will tell.
Qwen Office public beta: Qwen3.8 + QoderWork + MuleRun + Wukong in one, backed by DingTalk's enterprise clients — can it recover the share lost in July?
Google's last stand: new leadership in place, everyone waiting for Gemini 3.5 Pro. From the Arena leaderboard, Google's current highest-ranked model is only #13, behind the Claude family, Qwen3.8-Max (#9), and Kimi K3 (#12) — this is probably the first time in Google AI's history it has been squeezed out by both Chinese and American rivals at the same time.
"Same line: model capability determines the ceiling, but entry point, compute, and cost determine who keeps the traffic."
Giants are losing speed collectively on the Web, but the money hasn't disappeared — it just changed runways. Tencent's Q2 earnings data shows: capital expenditure payments reached 59.3 billion RMB, full-year guidance raised to 180-200 billion RMB, nearly doubling from 87.5 billion in 2025. This money directly crushed free cash flow — Q2 turned negative for the first time, at negative 13.8 billion. But Tencent made a special "ex-caliber" note: excluding AI-related compute prepayments, free cash flow was actually 37.6 billion. Meaning the core business's blood-making capacity is intact; profits are being actively consumed by AI investment. Advertising growth is also corroborating AI's empowerment value — marketing services revenue grew about 20% YoY, mainly benefiting from the explosion of AI smart ad platform AIM+. More noteworthy than these numbers is WeChat's next move — the earnings disclosed that Tencent recently launched small-scale gray testing of in-WeChat AI agents in the past few weeks, driven by customized models focused on user privacy, WeChat scenarios, and inference efficiency. The ultimate entry point of 1.3 billion MAU is charging. When WeChat AI agents scale, it may not be another ChatGPT, but an AI operating system embedded in the social graph — this is the deepest variable in Tencent's AI strategy.
Unique Research Rankings Release
Unique Research — Authoritative AI Market Insights Institution
Unique Research's July 2026 AI Web ranking details are as follows:
AI Website Traffic Chart: TOP 100 by visits and MAU for AI Web products, split into global and China markets.
All ranking charts also provide growth charts, helping investment institutions find promising projects and entrepreneurs and developers find potential opportunities.
Our influence and recognition:
Unique Research's data and insights have been widely trusted and cited by top domestic and international investment institutions, academic entities, and media, becoming an important benchmark for their market analysis, investment decisions, and trend judgment. Our partners and citers include (but are not limited to):
Securities firms: Sinolink Securities, China Securities, Huayuan Securities, Haitong Securities, etc.
Authoritative media: China Securities Journal, TMTPost, Jiemian News, etc.
We firmly believe that only data that withstands verification by industry core participants has lasting value.
We believe in the power of open source and openness. Unique Research hereby makes three commitments:
This AI product ranking is permanently open-source and free, with complete methodology and raw data supporting full reproduction.
This AI product ranking's data sources are strictly limited to independent third-party monitoring platforms; self-reported enterprise data is not accepted as evaluation basis.
This AI product ranking's ranking system maintains absolute neutrality, accepting no commercial cooperation or paid practices that interfere with ranking results.
How to use this ranking?
If you are an investor/analyst: this ranking provides you with a verifiable, clean data benchmark — a reliable foundation for building investment models, finding underwater projects, and cross-validation.
If you are an AI entrepreneur/product manager: you can not only precisely locate competitors, but also deeply reproduce our methodology to analyze your own product's market performance and optimize iteration.
If you are a researcher/student: the fully open-source nature makes this report an excellent free resource for academic research and market analysis. We encourage innovative research based on this data.
For complete data behind the charts and rankings shown in this article, please visit 100aiapps.cn. All visible data is copyable and open for free non-commercial use in personal study and research. Commercial use requires prior authorization consultation with Unique Research. We welcome listed teams to apply rankings to business plans, recruitment materials, etc., under compliant use conditions, providing authoritative evidence of your strength.
Join the AI Think Tank Open Source Community
Unique Research not only focuses on AI industry research, but also hopes to build an open-source community promoting AI market data transparency. We invite developers, data scientists, and industry analysts:
Reproduce and verify: visit our Feishu knowledge base for complete data and methodology, verify our results and suggest improvements.
Build together: contribute code to help us expand data dimensions (such as API call volume estimation, agent monitoring models, etc.).
Exchange insights: join our Feishu group to share unique insights based on Unique data.
One person's insight is limited; a group's wisdom is infinite. Let's define the measurement standards for the AI market together.
01. AI Website Traffic Chart
1.1 Global AI WEB TOP 100 by Monthly Visits
Based on Unique Research's independent AI industry research and continuous tracking of thousands of AI companies and nearly 10,000 AI products globally, sorted by that month's cumulative visits for each AI WEB product, organized into the global top 100 AI WEB products by visits.
1.2 Global AI WEB TOP 100 by Monthly Active Users
Based on the same research and tracking, sorted by that month's active user count, organized into the global top 100 AI WEB products by active users.
1.3 China AI WEB TOP 100 by Monthly Visits
Based on the same research and tracking, sorted by that month's cumulative visits for AI products serving the domestic Chinese market, organized into the top 100 AI WEB products by visits.
1.4 China AI WEB TOP 100 by Monthly Active Users
Based on the same research and tracking, sorted by that month's active user count for AI products serving the domestic Chinese market, organized into the top 100 AI WEB products by active users.
02. Research Statement
2.1 Data Description
The AI product revenue data in this report, article, ranking, or chart is based on real-time tracking of redirect traffic from AI product official websites to payment gateways (such as Stripe), combined with time-series analysis, deeply analyzing historical visits, monthly active user fluctuations, and pricing strategy changes to build a dynamic revenue evaluation model. Core model parameters are determined through a triple verification system: paid conversion funnel monitoring, enterprise-disclosed data reverse calibration, and industry benchmark conversion rate comparison, and timely cross-referenced and corrected with enterprise earnings, financing announcements, and authoritative media disclosures. Through sampling verification, the mean absolute error rate between predicted and actual data stably stays within ±10%, reaching industry research-grade accuracy.
The AI product traffic data in this report is based on multi-source data integration and Unique Research's intelligent algorithm processing, with monitoring scope strictly limited to users' direct access through official apps and websites. Specifically, this data only covers users' direct access to AI products through websites and native apps, and does not include traffic from: browser plugins/extensions, desktop client software, WeChat/Alipay mini-program ecosystems, Discord and other third-party embedded services, open-source model local deployments, API calls, and other non-direct access scenarios. This data focuses on AI product core-side access monitoring, aiming to objectively reflect mainstream user terminal direct usage.
Given the complexity and dynamism of data collection and processing, and continuously changing market conditions, the data shown may contain some degree of error and omission. Therefore, this data should be considered a research and analysis reference, aimed at providing users with a window into understanding the AI product market, not a deterministic basis for specific investment strategies or advisory recommendations.
Data coverage expansion note: We are actively developing and testing monitoring models for the following emerging areas, expected to be gradually included in future quarterly reports:
AI Agent ecosystem
AI hardware device sales and activity
Large model API call scale estimation
Stay tuned.
2.2 Concept Definitions
Geographic dimension
Overseas AI products: Products founded by non-Chinese entrepreneurs or teams and primarily launched for the global market (including but not limited to their home market).
Domestic AI products: Products founded by Chinese entrepreneurs or teams, primarily launched for the Chinese domestic market.
Going-overseas AI products: Products founded by Chinese entrepreneurs or teams but targeting overseas markets (non-Chinese domestic market).
Functional dimension
AI-native applications: Applications that deeply integrate AI technology and algorithms from the product design stage, whose core value, business processes, or user experience are entirely dependent on AI technology. These applications not only use AI to optimize or enhance existing features but treat AI as a core component — without AI technology, these applications would not exist or would lose their core value.
AI-feature applications: Applications that enhance existing features or provide new features by integrating or embedding AI technology on top of existing business logic and application frameworks. These applications may have already had mature business models and market positioning, but introducing AI technology can significantly improve efficiency, user experience, or create new value points.
AI ecosystem applications: Platform-type applications that provide support, connection, or communication facilitation for the AI ecosystem composed of AI technology developers, application developers, service providers, and end users. These applications go beyond the technical level, involving multiple dimensions such as market, community, and resource sharing, aiming to promote AI technology popularization, innovation, and collaboration.
2.3 Metric Definitions
WEB data metrics
Visits: The total number of all visits to a website within a specific time period (such as a month), measuring website traffic scale. Page views within continuous active periods count as the same session; users re-activating after 30+ minutes or starting a new day are counted as new visits.
Unique Visitors: The number of independent IPs visiting the target website within a specific time period (such as a month). If someone visits a website on multiple days in a month, they are only counted as one unique visitor.
ARR: Within a certain period (such as a year), the total revenue a website earns through subscription services provided to users, excluding advertising revenue, transaction commissions, and professional services.
APP data metrics
Downloads: The number of times an App is downloaded by users within a specific time period (such as a month), typically measured as downloads on an app store (such as App Store, Google Play).
Active Users: The number of independent users who performed at least one activity within a specific time period (such as a month).
IAP: Revenue generated by users purchasing virtual goods, additional services, or premium features within an App, excluding advertising revenue, direct user payments (such as tips), and third-party Android app store revenue.
2.4 Disclaimer
This report is published by Unique Research, and its copyright belongs to Unique Research. Any Chinese republication or citation must indicate the report source. Foreign institutions requiring republication or citation should contact for authorization in advance. When citing data, please label: [Data source: Unique Research]. When directly citing charts, please label: [Data source: Unique Research, Chart production: Unique Research].
This report is Unique Research's independent original analysis as a third-party institution. The content does not represent any enterprise's position and does not constitute investment advice to anyone. Investors should note that any investment decisions made based on this are unrelated to Unique Research and its employees or affiliated institutions.
To the extent permitted by law, Unique Research and its affiliated institutions may hold equity in companies mentioned in the report, or provide or seek to provide fundraising or financial advisory services, and its employees may serve as directors of companies mentioned in the report.