What it is in one line
360 bundling "multi-model plus multi-agent" into "one sentence generates video,
reports, and PPT" — selling "time saved on learning editing and making decks."
But "the agent swarm" currently reads as concept packaging; users cannot tell it
apart from ordinary multi-model switching.
Who built it
360 (601360), led by Zhou Hongyi, launched April 2025 as the consumer-facing
piece of 360's "AI + security" dual line. This entry
(纳米ai-首创多智能体蜂群) is its app; the pool also carries 纳米ai
(the web entry), both being two entry points of the same product "Nano AI," with
a third entry outside this batch. Official framing: connects 80+ mainstream
models, using agents to handle multimodal content generation, especially video.
Read: Nano AI is 360's all-in consumer bet on transformation, championed
personally by Zhou. The bet's landing point is "AI for everyone" — letting
people without professional skills produce commercial-grade content.
What it actually does
- One-sentence image/video generation → claims commercial-grade content in
60 seconds, with a built-in video editor
- Agent swarm → hundreds of specialized agents splitting composition, color,
copy, and sound design (official framing)
- Multi-model access → one app connecting multiple mainstream models
- Knowledge bases and custom agents → users can build knowledge bases and
customize their own agents
- Search/Q&A → quick answers plus deep reasoning
What it deliberately does not do: it does not rely on an in-house model —
core models come from external vendors, and 360's role is an orchestration and
packaging layer.
What old behavior it replaces
Making a short video used to require the full chain of script → shooting/editing
→ color grading → voiceover, meaning either learning editing software (high
learning cost) or hiring out (hundreds of yuan per piece). A deck took a full
day. Nano AI's pitch is "natural language as productivity" — compressing
professional content production from "learn tools, hire freelancers" into "say
one sentence."
For small merchants, self-media creators, and ordinary people doing content on
the side, it replaces the time and money cost of "outsourcing or self-teaching
editing." That is exactly where Zhou's "AI for everyone" narrative lands.
Business model
"Free trial plus light payment" (membership). No specific subscription pricing or
revenue disclosed. At the 360 report level, H1 2025 revenue was ¥3.827B with R&D
at 40.89% of revenue, but Nano AI revenue is not broken out.
Read: Nano AI is a strategic product, not a revenue product — 360 needs it to
carry the "AI transformation" story; making money comes later. The free-first,
light-payment structure confirms this.
Hard numbers
- MAU 8.08M, -4.46% MoM (traffic board, June 2026 global app ranking, #53)
- 360 official (Sept 2025): Nano AI app MAU over 12M; combined monthly web
visits over 450M across products, with "Nano AI Search" ranked #2 domestically
and "Nano AI Agent" #3
- Media figures (2025): 140M+ users served since launch, 300+ cities, DAU over
10M, 500K+ daily content generations — not cross-validated by third parties
- Third-party doubt: QuestMobile (June 2025) measured Nano AI Search app MAU at
only 1-5M, with a -8.6% compound growth rate — a huge gap from the
official 12M
- 360 summit framing (2025): 30M downloads, 400M monthly PC uses
Four-way read
| Dimension |
Call |
| Founder-product fit |
Zhou champions it personally; 360's transformation is genuine |
| Product insight |
"Agent swarm" is concept packaging; user-perceivable differentiation is weak |
| Execution quality |
Core models from external vendors; 360 is the orchestration layer, engineering adequate |
| Timing |
Video generation is crowded; 360 enters on channels and the "AI for everyone" narrative |
The call
Nano AI's app is a product whose data series contradict each other, and
judgment must anchor on third-party numbers. The official line is "12M+ MAU,
140M+ users, 10M+ DAU"; QuestMobile measured 1-5M MAU in the same period, and
traffic board measured 8.08M. Three series differ by an order of magnitude. When the
official series cannot be cross-validated, the trustworthy anchors are
third-party data — and they show a single-digit-million scale with negative
growth.
-4.46% MoM is a negative signal. Under any series, the app is declining
month over month, meaning the "agent swarm" selling point has not converted into
sustained growth. Read: multi-agent is becoming an arms-race term in Chinese
products — users cannot tell "swarm" from "multi-model switching," and this
packaging's marginal effect is fading.
Its real advantage is channels, not product. 360 owns PC search and browser
incumbency, and Nano AI is the AI-facing surface for that traffic — which is why
its "user scale" and "product heat" never line up: part of the users were
pushed there, not pulled by the product.
Worth watching strategically: in July 2026, 360 launched the enterprise-side
"Nano Work," throwing 100K+ agents into real business scenarios to iterate. If
consumer Nano AI loses resources to the enterprise line, app growth will look
worse.
What to watch next
① Whether app MAU stops declining next month — sustained negative growth means
the "swarm" story has not landed
② Whether the gap between "140M users / 12M MAU" and third-party numbers
closes — until the series are reconciled, the product's figures cannot be trusted
③ Whether consumer Nano AI loses resources to the Nano Work enterprise line —
decides whether it stays an entry point or gets downgraded to a demo
What you can take from it
Product logic: none. "Multi-agent swarm" is currently a marketing word, not a
product fact — build the user-perceivable difference first, name it later.
Doing it in the other order is concept-first.
Pricing structure: free trial plus light payment — a reasonable starting
structure for a product that still needs to educate the market, provided the free
tier actually creates retention.
Verdict
Unproven. The channel advantage is real, but product differentiation is weak,
app growth is negative, and official data clash badly with third-party series.
Until "agent swarm" shows genuine user-perceived value, or the series get
reconciled, it does not belong in a decision set.