Moonshot AI 月之暗面 (Kimi)
Company profile
Section titled “Company profile”- Founded: 2023-03 (founder Yang Zhilin 杨植麟, Tsinghua + CMU background, formerly an intern at Meta AI)
- Headquarters: Beijing
- Main product: Kimi (consumer / API)
- Model series:
- Kimi K1 (2024): 2-million-character super-long context, the breakout model
- Kimi K1.5 (2025-01-20): math / coding / multimodal reasoning at OpenAI o1 level
- Kimi K2 (2025-07): 1-trillion-parameter MoE / 32B active, open weights
- Kimi K2.5 (2026-01-27): the latest open flagship, 1T / 32B active
- Kimi-K2-Instruct-0905 (2025-09): 256K-token context
- Kimi Linear (linear attention; 128K–1M long context with 5–6× decoding speedup)
- Kimi K2.6 Code Preview (2026 Q2, forthcoming code / agent model)
- Kimi K3 (in development, targeting 3–4 trillion parameters)
- Business model: consumer subscription (Kimi app) + API + enterprise customisation
- Funding: $500M Series C led by IDG in 2025-12, with Tencent / Alibaba / Wang Huiwen 王慧文 following on; Hong Kong IPO under evaluation in 2026-03
Strategic positioning: long context + open-source catch-up + rapid monetisation
Section titled “Strategic positioning: long context + open-source catch-up + rapid monetisation”Built on long context
Section titled “Built on long context”Kimi K1 (2024) broke out in China with a 2-million-character Chinese context, cementing long context as Moonshot’s core technology signature:
- Kimi Linear pushes long context toward the technical frontier (productising linear attention)
- Long-document analysis and long-dialogue memory became Kimi’s distinctive capabilities
- Differentiation from Claude / GPT rests on the combination of Chinese-language strength and ultra-long context
Rollercoaster user base
Section titled “Rollercoaster user base”| Period | MAU | Notes |
|---|---|---|
| 2025 Q1 | 21.65 million | Peak |
| 2025 Q4 | 9.03 million | ~58% decline, share lost to Doubao / DeepSeek |
| 2026 Q1 | after Kimi K2.5 launch | Paid orders +8,280% m/m (2026-01), +123.8% (2026-02) |
Significance: Kimi’s user base fell but paid capacity surged — C-end ARR broke $100M (one month after K2.5 launch). A typical case of Chinese AI consumer products moving from traffic to value, and Kimi jumped from outside the top-100 to #9 on Stripe’s global ranking.
Open-source strategy
Section titled “Open-source strategy”Both Kimi K2 and K2.5 were fully open-sourced (HuggingFace):
- With Alibaba Qwen and DeepSeek, forms the “big three” of Chinese open source
- Relative to Qwen’s “ecosystem acquisition” and DeepSeek’s “maximalist open source,” Kimi leans toward technical brand
- Open source attracts international AI research attention → supports domestic funding and valuation
Deep dive: a commercialisation exemplar among Chinese AI-native startups
Section titled “Deep dive: a commercialisation exemplar among Chinese AI-native startups”Differentiation within the Chinese AI-native-startup cohort
Section titled “Differentiation within the Chinese AI-native-startup cohort”Since 2024, clear differentiation has emerged among Chinese AI-native startups:
| Company | Status as of 2026-04 | Primary narrative |
|---|---|---|
| ZhipuAI 智谱 | Listed on HKEX (2513.HK) | National team + academic heritage |
| Moonshot 月之暗面 | Hong Kong IPO prep + C-end ARR $100M | Long context + paid breakout |
| MiniMax | Overseas hit (Talkie) | Multimodal + going global |
| Baichuan 百川智能 | Pivot to healthcare vertical | B2B scenario-driven |
| StepFun 阶跃星辰 | Multimodal competition | Step-2 flagship |
| 01.AI 零一万物 | Business consolidation controversy since 2024 | Founded by Lee Kai-Fu 李开复 |
What makes Moonshot stand out:
- A single flagship product (Kimi) drives ARR growth
- Paid conversion higher than peers
- High C-end brand recognition
Self-regulation posture
Section titled “Self-regulation posture”Consistent with other Chinese companies, Moonshot has no standalone safety framework:
- Each Kimi version is filed with the CAC
- TC260-003-2024 compliance
- Kimi K2 / K2.5 technical reports include some safety-evaluation disclosure
- GPAI CoP not signed
But a planned Hong Kong IPO implies upgraded disclosure obligations — like Zhipu’s path, Moonshot in 2026–2027 may become an important observation site for Chinese AI governance transparency.
Compared with international peers
Section titled “Compared with international peers”- vs. Anthropic Claude: comparable long-context length (Claude 200K), but Kimi has a clear Chinese-language advantage
- vs. OpenAI: catching up on C-end paid capacity
- vs. Meta Llama / Mistral: similar open-source strategy, but focused on the Chinese-language market
Policy document snapshot
Section titled “Policy document snapshot”| Type | Document | Link | Subpage |
|---|---|---|---|
| Usage policy | Kimi user agreement + privacy policy | moonshot.cn / kimi.moonshot.cn | — |
| Technical reports | Kimi K1.5 / K2 / K2.5 Technical Reports | arXiv + GitHub | — |
| Open-source models | Kimi K2 / K2.5 GitHub | github.com/MoonshotAI | — |
| Transparency | CAC algorithm filings + (future HKEX disclosures) | cac.gov.cn | — |
Regulatory-compliance posture
Section titled “Regulatory-compliance posture”- China:
- Multiple CAC algorithm filings for Kimi versions
- TC260-003-2024 compliance
- Content-moderation headcount is smaller than peers (Kimi is a chat assistant, not a content-generation platform)
- United States:
- Kimi open weights downloadable in the U.S.
- No U.S. consumer services
- European Union:
- Same as above
- GPAI CoP not signed
- Japan / Southeast Asia: exploring overseas expansion since 2025
Company posture, in brief
Section titled “Company posture, in brief”- Technical-brand orientation: competitive focus on long context, agents, and coding
- Aggressive commercialisation: C-end paid breakout + Hong Kong IPO prep
- Open-source friendly: Kimi K2 series fully open-sourced
- Highly internationalised funding: $500M Series C (2025-12) signals a secondary-market path