Qwen's 58x Surge: Why Last Year's China AI Brand Benchmarks Are Obsolete in 2026

One number in QuestMobile's H1 2026 AI app report should change how international brands read their China AI visibility scores: Qwen (千问) grew its monthly active users by 5,792.9% year on year, to roughly 167 million. Doing the arithmetic backwards, that means Qwen had only about 2.8 million users a year earlier. Any brand benchmark, agency baseline or board slide that treated Qwen as a minor model in 2025 is now measuring the wrong thing. The model moved from rounding error to the second-largest independent AI app in China in twelve months, and brand visibility work has to be re-weighted accordingly.

The H1 2026 numbers

QuestMobile's H1 2026 AI application market report, covered by IT之家 and HK01, ranks standalone AI-native apps by June 2026 monthly actives. Pre-installed phone-maker assistants are excluded, so the figures reflect apps people chose to download.

The share column is our own calculation on the three leaders combined (about 679M MAU). The growth figures reported alongside are Doubao at +172.1% and Qwen at +5,792.9%. Working backwards, Doubao stood near 140M a year ago, so it added roughly 240M users while Qwen added roughly 164M from a near-zero base.

What a 58x jump does to a brand score

A brand's visibility score on a model is only as meaningful as the audience that model reaches. Two things happen when one model grows 58x in a year.

First, the weighting problem. If a brand's cross-model average treats six models equally, a 2025 Qwen score that reflected a 3M-user product carried the same weight as Doubao's. In 2026, Qwen's share of the three-leader audience is about a quarter. A brand that scores 80 on Doubao and 20 on Qwen is in a very different position from one scoring 20 on Doubao and 80 on Qwen, even though the simple average is identical. Our earlier analysis of the Qwen surge and Kimi's fall flagged the direction of travel; the H1 data now puts a hard number on it.

Second, the baseline problem. Brands that ran a GEO audit in late 2025 and filed it as "the China AI picture" are working from a snapshot in which Qwen was effectively unmeasured. Content decisions made on that snapshot, such as which platforms to seed, which entity pages to maintain and which citations to chase, were optimized for an audience mix that no longer exists.

Why Qwen's growth matters more for brands than the headline suggests

Qwen's growth is not just a volume story. It sits inside the Alibaba ecosystem, and as we covered in Quark and Tmall's citation advantage, Alibaba-owned surfaces tend to favor commerce-linked sources such as Tmall product pages and brand store content. A model that grew from 3M to 167M users with a commerce-heavy citation habit is a very different channel from one that leans on Douyin content or open-web articles.

For brands this means the same query can now resolve to three different evidence sets:

• Doubao (382M): leans on ByteDance content surfaces, so Douyin and Toutiao material is disproportionately visible.
• Qwen (167M): leans on the Alibaba commerce graph, so product-data completeness and store signals matter.
• DeepSeek (130M): leans on open-web retrieval and its parametric knowledge, so encyclopedic and authoritative third-party sources carry weight.

A brand strong in only one of those evidence sets is exposed in the other two, and the audience of the "other two" is now roughly 44% of top-three volume when you stand in Doubao's shoes, or 81% when you stand in Qwen's.

Three scenarios for a brand scoring well on Doubao but poorly on Qwen

The gap between models is where re-baselining pays off fastest. Consider three common patterns we see in international brand audits.

The Douyin-strong brand. A fashion or beauty label with heavy Douyin and Xiaohongshu seeding often performs well on Doubao, because ByteDance surfaces feed that model. But if its Tmall flagship listings are thin, with short titles, missing specifications and inconsistent naming against the official site, Qwen has little commerce-grade evidence to cite. With Qwen now at 167M users, that weakness reaches an audience larger than the entire 2025 Chinese AI-native user base of most competitors' targeting plans.

The authority-strong brand. A premium or technical brand with deep Baidu Baike entries, trade-press coverage and encyclopedic references tends to do better on DeepSeek, where open-web authority and long sessions favor depth. Its risk is the opposite: it may be invisible in the short, commerce-flavored answers that Qwen produces inside Alibaba's shopping surfaces.

The newcomer. A brand entering China in 2026 has an advantage in one respect: it has no stale 2025 baseline to defend. It can build entity consistency across all three evidence sets from day one, rather than retrofitting. The cost is that it must do so across three ecosystems at once, which is exactly why a per-model reporting split matters.

How to read the growth figures without overreacting

A 5,792.9% growth rate is arithmetically dramatic because the base was tiny, and brand teams should resist two opposite mistakes.

The first mistake is dismissing it as a promotional spike. Qwen's growth came alongside heavy consumer promotion and ecosystem integration, which our coverage of the Spring Festival subsidy war examined. Subsidy-driven users can churn. But even if a meaningful share does, the retained base is far above the 2025 level, and the citation behavior of the model is what brands compete on regardless of user mix.

The second mistake is extrapolating the curve. Qwen will not grow another 58x. The sensible planning assumption is that audience shares among the top three will keep shifting by single-digit points per quarter, not by multiples. That argues for quarterly re-weighting rather than a one-off panic rebuild. Doubao's own June gain of 13.78M users in a single month is a reminder that even the leader is still moving.

What this means for agencies and vendors

For China-market agencies, the Qwen number is also a reporting-integrity issue. Any client deck that shows a single blended "China AI visibility" figure without disclosing model weights is hard to audit, and a year-old weighting scheme can quietly overstate or understate performance. Clients should ask vendors three questions: which models are included, what weight each carries, and when the weights were last updated. A vendor that cannot answer the third question is probably still using a 2025 map. Our analysis of score integrity under model version churn covers the related risk that scores drift even when your content has not changed.

The engagement caveat

Raw MAU is not the whole picture. The same report shows the share of users with 10-plus minute sessions: DeepSeek 30.0%, Doubao 27.5%, Kimi 26.1%. DeepSeek's audience is smaller than Qwen's but spends longer per session, which is the kind of behavior associated with research-style, multi-step queries where brand shortlists get formed. Native AI apps overall average 183 minutes of monthly use, up 40% year on year.

The practical reading: Qwen is the new reach channel, DeepSeek is the deliberation channel, and Doubao is both the largest and the most contested. Our piece on the MAU-versus-engagement budget trap makes the same point from the budgeting side; the Qwen growth number adds the urgency.

A re-baselining checklist

Brands should not need a new strategy, but they do need to refresh the measurement under the existing one:

1. Re-run every 2025 audit on Qwen specifically. Treat any Qwen number older than six months as missing data.
2. Weight by audience, not by model count. Use June 2026 MAU as a first-pass weight and revisit quarterly, since Doubao alone added 13.78M users month on month in June.
3. Split reporting by evidence set. Report visibility separately for ByteDance-fed, Alibaba-fed and open-web-fed models rather than one blended score.
4. Audit commerce data for Qwen. Check product titles, specifications, pricing and store content on Tmall and related Alibaba surfaces for completeness and consistency with your official pages.
5. Keep a DeepSeek deliberation layer. Maintain authoritative, citable long-form sources, because longer sessions reward depth.

Takeaway for Brand Marketers

The lesson of Qwen's 58x year is that China AI audience shares are not stable enough to benchmark once a year. If your last full visibility audit predates 2026, your Qwen picture is effectively blank, and Qwen now reaches around a quarter of the top-three audience. Re-baseline this quarter, weight scores by actual audience, and stop reporting a single blended China AI number. Brands that do this will spot gaps in the Alibaba-fed channel while competitors are still optimizing for the 2025 map. Also treat Kimi's 7.3M MAU as a niche, high-intent channel rather than a core pillar; it is a fraction of Qwen's audience.

Caveats: QuestMobile's ranking excludes pre-installed phone assistants, so total exposure for models bundled on handsets is higher than shown, and the growth percentages come from the vendor's own year-on-year comparison. Backed-out prior-year figures are our own arithmetic, not reported numbers.

Related: Brand visibility scores by model on /brands · Qwen surge, Kimi fall · MAU vs engagement budget trap