Summary
MiniMax — the Chinese AI startup listed on HKEx (4 months ago, up 78% on day one) — just disclosed: 100K+ enterprise and developer customers (5x growth in 6 months), ~300M global users, ARR doubling every 60 days. Only Anthropic is faster. >70% revenue comes from outside China. This is the story of how a company with 1/10 the headcount and 1/100 the capital of OpenAI is winning the agent era.
The Numbers
| Metric | Value | Context |
|---|---|---|
| Developer customers | 100K+ | 5x in 6 months |
| Global users | ~300M | — |
| ARR doubling cycle | 60 days | Only Anthropic is faster (~4 months) |
| Team size | ~2,000 | OpenAI has ~20,000 |
| Capital efficiency | 1/100 of OpenAI | — |
| Overseas revenue | >70% | — |
| Active agents | 50K+ | Running on platform |
| Listing | HKEx, Jan 2026 | Up 78% on day one |
The Timeline
- Late 2021: Founder Yan Junjie leaves SenseTime to start MiniMax
- 2025 H2: Launches open platform for developers
- Jan 9, 2026: HKEx listing (market cap ~$90B HKD)
- Feb 2026: M2.5 model — top-tier at coding and tool calling
- Mar 2026: M2.7 (agent-optimized), adapts OpenClaw → MaxClaw
- Apr 2026: MaxHermes (first cloud self-evolving AI agent)
- May 2026: Mavis (multi-agent team collaboration)
- May 28, 2026: Discloses 100K+ developer customers
Why Developers Love MiniMax
Three things matter to developers:
- Model quality — M2.5/M2.7 rank at or near the top for tool calling, coding, and agent orchestration
- Price — API pricing is ~1/3 of international peers
- Toolchain — End-to-end: training → fine-tuning → deployment → monitoring
MiniMax delivers all three. And the network effect is real — once a team builds their agent pipeline on MiniMax, switching costs are high.
The Strategy
MiniMax's playbook sounds familiar if you follow Anthropic:
- "Small cut, deep penetration" — Don't compete on parameters. Compete on developer experience.
- Agents first — They saw the agent era coming and built for it. 50K+ agents running on their platform.
- Global from day one — >70% overseas revenue. They didn't build for China first.
- Capital efficiency — 1/10 the team, 1/100 the capital. Every dollar matters.
What This Means
The LLM landscape is shifting:
- Phase 1 (2023-2024): "Bigger is better" — parameter count was king
- Phase 2 (2025-2026): "Better is better" — quality of interaction matters
- Phase 3 (2026+): "Useful is better" — agents, tools, developer ecosystem
MiniMax skipped Phase 1. They went straight for Phase 3. And it's working.
FAQ
Q: Is MiniMax available outside China?
A: Yes. >70% revenue is overseas. Their API works globally.
Q: How does MiniMax compare to DeepSeek?
A: Different focus. DeepSeek is research-first (open weights). MiniMax is developer-first (API + agents).
Q: Can I use their models today?
A: Yes. Their open platform is live at platform.minimaxi.com.
The Chinese Anthropic is real, it's public, and it's growing fast. Watch this space.
Top comments (0)