Yuanbao Has Two MAU Numbers, 2x Apart: Which One Should Set Your China AI Brand Budget in 2026?

Two credible-looking sources describe Tencent's Yuanbao in the same summer and land on figures that differ by a factor of two. A 36Kr analysis published on August 3, 2026 says Yuanbao "regained more than 100 million monthly active users in April." QuestMobile's 2026 H1 AI-native app ranking, reported on August 4–5, puts Yuanbao's June MAU at 49,841,100. For a brand team splitting a China GEO budget, that gap decides whether Yuanbao is a clear fourth-place channel or a rounding error. This piece walks through what each number most likely measures and how to budget when you cannot be sure.

The QuestMobile baseline: a three-app market and a long tail

QuestMobile's H1 2026 ranking (June 2026 data) is the cleanest public snapshot of China's AI-native apps:

Shares are our calculation from the reported MAU figures; the five apps sum to roughly 758M.

Three things stand out. Doubao is about 2.3x Qwen and bigger than the next four apps combined. The top three together hold roughly 679M MAU, close to nine-tenths of the top-five pool. And the fourth slot is a cliff: Yuanbao sits at about 7.5% of Doubao's size on this measure. Kimi and Wenxin do not appear in the top five at all, which is itself a signal for brand teams that still treat all six major Chinese models as equal-weight channels.

The 36Kr claim: what "100 million" is describing

36Kr's Aug 3 piece, an opinion essay on why Qwen and Yuanbao seemed quiet during the summer model race, cites QuestMobile for Doubao (382M, +172.1% YoY) and Qwen (167M, +5,792.9% YoY, rank up from fifth to second). For Yuanbao it uses a different reference point: the app "regained more than 100 million" MAU in April. The same article notes that Tencent's Hunyuan Hy3 launched July 6 with a free agent feature built into Yuanbao, and that JD.com's AI agent plugged into Yuanbao's mini-program ecosystem on July 15.

We cannot tell from the public text which counting method produced the 100M figure. Plausible explanations, all hypotheses rather than confirmed facts: (1) a different month, since April and June are different windows and Yuanbao's traffic is promotion-driven; (2) a different definition, such as counting entry points inside WeChat or other Tencent surfaces versus QuestMobile's standalone-app view; (3) a different source, such as company-reported figures versus a third-party panel. The pattern is common in China's AI market: first-party numbers tend to be broader, panel numbers narrower.

Why the definition matters more for Yuanbao than for anyone else

Doubao, Qwen and DeepSeek are mostly destination apps. Users open them to ask questions, so app MAU is a reasonable proxy for query volume. Yuanbao is different because its strategic value sits in Tencent's distribution: WeChat, mini-programs, and now commerce agents such as the JD.com integration. If a meaningful share of Yuanbao-powered answers is delivered through those surfaces, a standalone-app panel undercounts exposure, and the 100M-plus figure may be closer to what a brand actually faces. If most of those users are casual or promotion-driven, the 49.8M figure is the better planning number.

Here is the budget consequence. At 49.8M, Doubao is about 7.7x Yuanbao. At 100M it is about 3.8x. Those two ratios justify completely different allocations, which is why picking a single number without understanding it is risky.

A practical way to budget when the data disagrees

We would not try to resolve the argument; we would size Yuanbao so the decision survives either answer.

Treat the low number as the floor and the high number as the ceiling. A defensible Yuanbao allocation is the one you would still be comfortable with if the true audience were 50M, and that you would regret not making if it were 100M. In practice that means baseline monitoring plus fixes to the high-leverage entity facts (brand name variants, official descriptions, product specs), not a bespoke content program.

Test the surface, not the app. Run your brand-category queries inside Yuanbao in both the standalone app and WeChat-integrated entry points, and log whether the brands mentioned differ. If they do, you have evidence that "Yuanbao" is two measurement targets, and your dashboard should say so. This is the same logic behind testing multiple Qwen surfaces, which we covered in our Qwen3.8-Max open-weights analysis.

Weight by category, not by headline MAU. Yuanbao's JD.com tie-in matters most for shopping-led categories; a B2B software brand has less reason to chase it. Our category benchmarks on the /brands page show how scores differ by industry and model, which is a better allocation input than any single MAU figure.

Re-check each quarter. Yuanbao's number has moved sharply twice this year by the sources' own accounts (a regain in April, a Hunyuan Hy3 launch in July). QuestMobile's next release will show whether the July integrations changed the picture.

What this says about reading China AI statistics generally

The Yuanbao discrepancy is a reminder to write down the source, month and definition next to every MAU figure in a GEO deck. Of the three numbers that matter most this year, only QuestMobile's June snapshot is both dated and specific: Doubao 382M, Qwen 167M, DeepSeek 130M. Claims about fourth place and below deserve a footnote. A brand that allocates budget off a rounded "100 million" in a headline may be overpaying by a factor of two, and one that dismisses Yuanbao off the 49.8M panel number may be ignoring the channel Tencent is working hardest to embed in WeChat.

Three signals to watch before the next ranking lands

Because the numbers are contested, it helps to track leading indicators that are visible to any brand team without a panel subscription.

Answer overlap with Doubao and Qwen. If Yuanbao increasingly mirrors the brand lists returned by the two bigger apps, its marginal value to you is low and one set of fixes covers all three. If its lists diverge, which our earlier cross-model analyses suggest is common when models rely on different evidence layers, it deserves its own checks. Measure this with a fixed set of 20 to 30 category queries run on the same day each month.

Source types cited in answers. Yuanbao sits inside Tencent's content ecosystem, so watch whether answers lean on WeChat official-account articles and Tencent properties rather than the open web. That determines where your content investment should go, regardless of whether the audience is 50M or 100M.

Commerce handoffs. The July 15 JD.com agent integration means some Yuanbao queries now end in a product page or an order, not just a text answer. Track whether your brand appears in those handoffs, because a mention that leads to a purchase flow is worth more than the same mention in a plain answer.

The cost of getting it wrong in either direction

Overweighting is the more visible risk: a team reads the 100M-plus claim, builds Yuanbao-specific content, and finds that the audience it reaches is a fraction of what was promised. Underweighting is quieter but can be more expensive. If Tencent's integrations push Yuanbao answers into WeChat surfaces where hundreds of millions of people already spend their day, a brand with no entity hygiene there will simply be absent when the traffic arrives. The floor-and-ceiling approach above is cheap insurance against both outcomes: most of the work is fixing facts that also help your Doubao, Qwen and DeepSeek visibility.

Takeaway for Brand Marketers

1. Anchor on the three-model core. Doubao, Qwen and DeepSeek cover roughly nine-tenths of the top-five MAU pool. Make sure your brand facts are right and your visibility is monitored there first.
2. Size Yuanbao for the range, not the headline. Budget for the 50M floor, keep an option open for the 100M ceiling, and decide by category (commerce-led brands lean higher).
3. Footnote every MAU figure. Record who published it, which month, and whether it is standalone-app or ecosystem-wide. Do not mix sources in one chart.
4. Test where the answer is delivered. Query Yuanbao through its standalone and WeChat-adjacent entry points and compare which brands appear.
5. Revisit after the next QuestMobile release. If Yuanbao's panel figure moves materially toward 100M, raise its allocation; if not, keep it at monitoring level.

Sources: QuestMobile 2026 H1 AI-native app ranking via Tencent News (Aug 5, 2026) and Phoenix News (Aug 4, 2026); 36Kr analysis of Qwen and Yuanbao (Aug 3, 2026). Share and ratio figures are hubGEO calculations.

Related: Brand scores by model on /brands · The Hunyuan Blind Spot · China's AI Fourth-Place Cliff