Web traffic is evolving from "a scale metric for AI products" into "an entry-point structure metric for AI products."
As usual, let's open with three numbers.
**+552%**: Alibaba's "Qwen Office" launched public beta on August 3. In its first full month, it claimed the #1 spot on China's growth ranking.
**+135%**: Tencent's WorkBuddy visits more than doubled again — its second consecutive month of doubling.
**1 billion**: Google officially announced Gemini App monthly active users surpassed 1 billion. But in the same month, Web-side visits dropped 2.6%.
The first two numbers define the hottest track in China's AI circle in August.
The third number changes how this ranking should be read going forward.
Below is Unique Research's compiled global and China AI Web data for August 2026.
---
Let's start with that 1 billion.
On August 11, Google officially announced: Gemini App MAU exceeded 1 billion, with 63% of users using voice interaction, and the Android client already capable of automatically operating across 40+ apps.
Throughout August, Google's product moves pointed in one direction — stuffing Gemini into phones, Chrome, devices, and every kind of system entry point.
So gemini.google.com dropping 2.6% absolutely does not mean Gemini users dropped 2.6%.
Quite the opposite: it may signal that AI usage is breaking away from the "open an AI website" paradigm.
Claude is the same story. In August, Claude website visits were -1.97%, but MAU actually grew +3.16%.
Because everything Anthropic did this month happened outside the website.
On August 12, it upgraded the Chrome sidebar into a full Cowork work session. Around the 25th, it enabled cross-device memory. On the 27th, it stuffed a built-in browser into Cowork that can directly import your Chrome login state.
You never leave the browser, and Claude has already finished the work.
So starting this month, we're adding a "reading note" to this ranking.
**Web traffic is evolving from "a scale metric for AI products" into "an entry-point structure metric for AI products."**
In 2025, you could tell who was hot just by looking at websites.
In 2026, users may be invoking the same AI from desktop clients, browser sidebars, IDEs, terminals, enterprise IM, and phone system entry points — while rarely visiting the website at all.
This ranking still has value, but what it measures increasingly is — who still needs users to actively open a website.
With that premise, let's look at August data, and many "anomalies" start making sense.
---
The protagonist of August's China ranking is a category of products: office agents that get work done for you.
"Qwen Office" launched public beta on August 3, consolidated from three internal products — QoderWork, MuleRun, and Wukong. Powered by Qwen3.8, it's the industry's first agent supporting desktop, cloud, and enterprise collaboration simultaneously, personally led by DingTalk CEO Chen Yusen.
On beta launch day, Alibaba's Hong Kong stock rose 7%.
Its cold-start playbook is textbook: 2,000 credits free at beta launch, plus 500–2,000 more for daily login in the first week.
On August 18, it officially integrated with WeCom (Enterprise WeChat), completing the trifecta of China's three major office platforms — DingTalk, Feishu, and WeCom — and顺手 also connected Slack.
First full month results: 1.09 million visits, +552% MoM, #1 in China growth ranking; 410,000 MAU.
WorkBuddy-CN: 14.09 million visits, +134.98% MoM; 3.76 million MAU, +123.21% MoM — its second consecutive month of more-than-doubling growth, firmly holding the top spot among China's office agents.
Its approach is very "Tencent": deep integration with WeCom, Tencent Docs, and QQ Mail. The model layer is agnostic — Hunyuan, DeepSeek, Zhipu GLM, and Kimi are all callable.
The overseas version workbuddy.ai is also gaining traction: 850,000 visits, +277% MoM.
Zhipu's ZCode is similarly steady: 2.33 million visits +27.75%, 1.25 million MAU +30.57%, holding ground at the intersection of coding and office.
This track had an industry-level backdrop in August.
On August 18, WeCom 5.0.10 fully opened CLI and MCP capabilities, exposing ten major office capabilities — messaging, docs, sheets, calendar, meetings, and more — to agents like WorkBuddy and Qwen Office all at once.
The walls around office software came down. Agents can now directly enter systems to fetch context, call tools, and close loops on work.
Qwen Office announcing WeCom integration the same day was no coincidence — it's the direction of the tide.
In the same month: Coze (扣子) visits -27.64%, codebuddy-CN -25.88%, Xiaohuanxiong (小浣熊) -18.93%, OpenClaw -18.08%.
Big-tech integrated products are surging, early single-product players are declining — models, entry points, and ecosystem resources are all concentrating toward the "regular army."
ByteDance merging Feishu into Doubao and launching Doubao Virtual Desktop for Windows (which can watch the screen, click the mouse, and type like you), Baidu merging its AI office product dodo into "Baidu Dazi" — all follow the same logic.
The office agent war: big tech isn't holding a horse race anymore — they're placing heavy bets directly.
Our own ranking category quietly changed this month from "lobster agents" to "office agents" — the lobster steps down, getting-work-done steps up.
---
- **ChatGPT**: 5.638 billion monthly visits, +3.45% MoM; 475 million MAU, +1.19%
- **Gemini**: 2.617 billion monthly visits, -2.59%; 303 million MAU, +0.36%
- **Claude**: 950 million monthly visits, -1.97%; 111 million MAU, +3.16%
- **Grok**: 210 million monthly visits, -0.39%; 30.99 million MAU, +0.22%
- **Perplexity**: 95.24 million monthly visits, -8.24%; 17.69 million MAU, -5.46%
ChatGPT visits hit a new high, with volume and users growing in sync — its base is terrifyingly stable.
Gemini and Claude's "website down, actual up" was covered at the opening — no need to repeat.
Gemini 3.5 Pro has been repeatedly delayed. In August, SemiAnalysis directly stated the model has been cancelled, and Brin was brought back to oversee.
Google's current state: application layer charging ahead, flagship model layer in a gap, relying on the Flash series to hold the line.
Even version numbers are misaligned: Flash has reached 3.8, while Pro is still stuck at 3.5 unreleased.
Perplexity has declined for multiple consecutive months — the halo of AI search is fading.
Grok is treading water. Grok Bot is currently only in testing on macOS and iOS; the Web side hasn't gotten a sip.
---
Kimi August visits: 51.24 million, -14.75% MoM; MAU 11.32 million, -26.35% MoM.
The pullback isn't surprising: K3 launch heat naturally subsided, C-end new user subscriptions were paused until mid-August — a full month of new additions choked off — and there was no new model to接力 in August.
What's worth noting is where it landed after the pullback — 11.32 million MAU, still substantially above pre-launch levels.
Launch pulses fade, but the base genuinely lifted. That's probably the confidence behind Moonshot AI pushing toward a $50 billion pre-money valuation and sprinting toward a Hong Kong IPO.
The control group is DeepSeek: chat main site basically flat in August (-1.41%), but open platform visits surged 45.29%, MAU +40.83%.
The V4 series launched consecutively in August, and the new model's gains landed squarely at the developer entry point.
On August 16, V4 even raised prices (peak-valley billing), but at $1.32 per million output tokens during peak hours — compared to K3's $15 — it's still a "price butcher."
---
The double #1 on both global growth and MAU growth rankings in August is a new name: OrcaRouter, 830,000 visits +931%, 490,000 MAU +1182%.
What does it do? Model routing — same track as OpenRouter.
And the timing of its surge couldn't be more perfect: on August 19, Stripe officially announced acquiring OpenRouter for over $7 billion. This company, valued at only $1.3 billion in its May Series B, saw its value quintuple in 3 months.
"Model routing" overnight went from a developer tool to infrastructure in the eyes of a payment giant — and alternatives in the same track naturally rode the coattails.
Another worth mentioning: Manus.
On August 11, it officially split from Meta and regained independence. Visits 25 million +27.73%, MAU 7.29 million +26.25% — returning to pre-acquisition-scandal levels.
The recovery is real, but the soul-searching question remains.
In an era where model vendors themselves are building agents, where is the middle layer's moat?
One month of data can't answer that.
The remaining spots on the growth ranking are basically divided among three categories.
- **Version update pulses**: ACE Studio 2.0, SubEasy, Eklipse — big version launches drive a spike
- **Back-to-school season demand**: Class Companion, Fermion and other education tools
- **Old domain, new project**: e.g., Rewind AI — after the original project was acquired and shut down by Meta, a new team启用 the domain in March this year to run a free AI tool site, eating the old domain's traffic base
When reading the growth ranking, all three categories should be discounted.
---
ERNIE rose another 92.79% in August (56.57 million visits, 16.14 million MAU) — looks scary.
But let's add up Baidu's three domains for the full picture:
- chat.baidu.com down to only 3.09 million (-89.67%), yiyan.baidu.com already fell out of the ranking — migration basically complete
- Three-domain combined visits ~59.66 million, basically flat vs last month
- Combined MAU ~17.62 million, almost exactly back to pre-migration levels
After two months of折腾:
Baidu AI's Web-side real plate hasn't gotten bigger — only the street address changed.
The real signal is beyond the data: on the August 18 earnings call, Robin Li declared "let ERNIE return to the first tier of foundation models," while Baidu's Q2 net profit attributable to shareholders fell 68% YoY.
Baidu can't afford to lose the ERNIE battle.
---
Regular readers know our filter: 5M+ MAU, double-digit MoM growth, regardless of background or buzz.
In August, this club expanded to 15 (ERNIE excluded due to methodology issues).
| Product | Category | MAU | MoM |
|---|---|---|---|
| Wanderboat | Personal assistant | 9.27M | +72.3% |
| Rodin | 3D modeling | 5.24M | +54.7% |
| DeepSeek Open Platform | Dev tools | 7.37M | +40.8% |
| DaVinci | Image generation | 5.58M | +27.5% |
| Manus | Agent | 7.29M | +26.2% |
| Meshy | 3D modeling | 6.87M | +24.3% |
| Quizlet | Education | 13.64M | +18.7% |
| Grammarly | Writing | 23.89M | +15.0% |
| GPTZero | Content detection | 5.16M | +14.4% |
| OpenRouter | Model aggregation | 6.21M | +13.8% |
| ZeroGPT | Content detection | 9.67M | +11.5% |
| Google Labs | Navigation | 13.14M | +10.4% |
| Qwen | Chatbot | 5.70M | +10.4% |
| ElevenLabs | Text-to-speech | 14.91M | +10.3% |
Three clues hidden in the list.
Quizlet (+18.7%), Grammarly (+15.0%) rebounded. ZeroGPT and GPTZero — two AI detection tools — both grew double digits simultaneously. Students are back writing homework, and teachers are back checking it.
Education and writing detection are the most reliable "seasonal business" around September every year.
Rodin MAU +54.7%, Meshy +24.3%. AI-generated 3D assets are evolving from toys into productivity tools for gaming and ecommerce.
Higgsfield MAU growth slowed to +5.7%. Kling, HeyGen, VEED all turned negative.
candy.ai only +1.7%, Character.AI -5.7%, SpicyChat -17.3%.
Hot topics rotate. Only "getting work done" and "back to school" are hard demand.
Also worth noting: Qwen's overseas version (chat.qwen.ai) MAU +10.4% — Qwen3.8's gains are starting to materialize overseas.
China's main Qwen site is still slightly down -2.92% — Alibaba seems genuinely uninterested in fighting the C-end chatbox battle on Web, focusing all energy on office agents.
---
- **Gemini 3.5 Pro — will it ever launch?** Cancellation rumors and "coming soon" leaks fly together. Whether it launches or not, Google owes the market an answer.
- **Qwen Office's second full month**: After the free-credit launch heat fades, how much of that 1.09 million will stick?
- **Kimi's Hong Kong IPO**: Filing rumors point to late September. What will the first post-IPO report card look like?
- **Office agents' "Three Kingdoms"**: Alibaba bets on DingTalk, Tencent on WeCom, ByteDance on Feishu. August was just the opening.
- **Stripe's OpenRouter acquisition aftermath**: After model routing becomes a giant's business, can neutrality be preserved?
August's data tells one story: the battlefield for AI products is spreading from "websites" to every entry point.
Office agents are grabbing desks inside enterprise IM. Gemini is grabbing entry points in phone systems. Claude is grabbing context in browser sidebars.
The Web ranking can still see the tip of the iceberg — but the part below the surface is growing larger and larger.
See you in a month for the September ranking.
---
Unique Research — authoritative institution for global AI market insights.
Unique Research's August 2026 AI Web ranking details:
- **AI Website Traffic Ranking**: TOP100 ranking of AI Web products by visits and MAU, distinguishing global and China markets.
- All rankings also provide growth rankings, helping investment institutions discover potential projects and entrepreneurs/developers discover opportunities.
For complete rankings and chart data, visit uniqueresearch.cn.
Unique Research's data and insights have earned widespread trust and citation from top investment institutions, academic units, and media domestically and internationally, becoming an important benchmark for market analysis, investment decisions, and trend research.
Our partners and citing parties include (but are not limited to):
- **Securities & Investment Banking**: Guojin Securities, CITIC Construction Investment, Huayuan Securities, Haitong Securities, etc.
- **Authoritative Media**: China Securities Journal, TMTPost, Jiemian News, etc.
We firmly believe that only data that withstands verification by core industry participants truly possesses long-term value.
We believe in the power of open source and openness. Unique Research hereby makes the following commitments:
1. This AI product ranking is permanently open source and free. Complete methodology and raw data support full reproduction.
2. Data sources are strictly limited to independent third-party monitoring platforms. Self-reported enterprise data is not accepted as an evaluation basis.
3. The ranking system maintains absolute neutrality. No commercial cooperation or paid behavior that interferes with ranking results is accepted.
- **If you're an investor/analyst**: This ranking provides a clean data benchmark that withstands reverse-engineering verification — a reliable foundation for building investment models, discovering underwater projects, and cross-validation.
- **If you're 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're a researcher/student**: The fully open-source nature makes this report an excellent free material library for academic research and market analysis. We encourage innovative research based on this data.
For complete data behind all charts and rankings shown herein, visit uniqueresearch.cn.
All visible data is reproducible and freely available for personal learning, research, and other non-commercial use. For commercial use, please consult Unique Research for authorization in advance.
We welcome listed teams to apply rankings to business plans, recruitment materials, and other scenarios under compliant use, providing authoritative corroboration for your strength.
Unique Research not only focuses on AI industry research but also looks forward to building an open source community that promotes AI market data transparency.
We invite developers, data scientists, and industry analysts to:
- **Reproduce and verify**: Visit our Feishu knowledge base to obtain complete data and methodology, verify our results, and propose improvements.
- **Co-build**: Contribute code to help us expand data dimensions (e.g., API call volume estimation, agent monitoring models, etc.).
- **Exchange insights**: Join our Feishu group to share unique insights derived from Unique Research data with peers.
One person's insight is limited; a group's wisdom is infinite. Let's define the measurement standards for the AI market together.