I’m looking to grow my twitter engagement and just social media presence in general. Sick of just endlessly scrolling through twitter, liking/retweeting other people tweets. I want to create content and show people what I can do…
Who know’s what I can become
Two Chinese AI equity perps just went live on Hyperliquid — $ZHIPU and $MINIMAX — and I took positions in both. Here's my thinking.
What these actually are
Both are HIP-3 equity perps launched by @tradexyz, tied to their Hong Kong listings (Zhipu = HK2513, MiniMax = HK0100). Before this, getting exposure meant opening a HK brokerage account, trading during HK hours, and holding spot only. Now it's levered, 24/7, permissionless, and onchain.
Zhipu (HK2513)
Spun out of Tsinghua in 2019, it's arguably the first pure-play AGI lab to go public. It listed January 8th at HK$116.20 and has since run roughly 20-25x, putting the market cap somewhere in the $120–150B range — though with a free float under 6%, daily swings of 25-30% are normal.
The real catalyst is GLM-5.2, a 753B-parameter MoE (around 40B active parameters) with a 1M context window, released mid-June under an MIT license. On coding benchmarks it sits ahead of GPT-5.5 on SWE-bench Pro and within a point of Claude Opus 4.8 on FrontierSWE — at roughly one-sixth the cost. For teams outside the US who just lost access to closed frontier models through export controls, this is about as good a substitute as exists right now.
The catch: 2025 revenue was only ~$100M against a ~$650M net loss. North of 1,000x sales is a valuation priced entirely on capability and scarcity, not fundamentals.
MiniMax (HK0100)
Founded in 2022 by an ex-SenseTime team and backed by Alibaba, Tencent, and miHoYo. It listed January 9th at HK$165, doubled on debut (the public tranche was oversubscribed nearly 1,800x), peaked around HK$1,330 in March, and has since pulled back to roughly a $20B market cap.
Their open-weight flagship M2.7 is the sleeper here. It matches GPT-5.3-Codex on SWE-Pro, hits 78% on SWE-bench Verified, and tops the GDPval-AA leaderboard among open-weight models — all at 50–60x cheaper per token than Claude Opus. One independent comparison put it at ~90% of Opus 4.6's output quality at around 7% of the cost.
Financials are similar story: ~$79M in 2025 revenue, ARR tracking above $300M with 300M+ users (70% of revenue from overseas), offset by a ~$250M adjusted loss. At ~$20B that's roughly 250x sales.
The bigger picture
Together, Zhipu and MiniMax are valued at somewhere between $140–175B combined. For comparison: Anthropic is privately valued at ~$965B, OpenAI at ~$852B. Two Chinese open-weight labs, shipping models within a few benchmark points of the US closed frontier, trade at less than a fifth of Anthropic alone. On a capability-per-dollar-of-market-cap basis, there's nothing cheaper right now — and as of today you can express that view onchain with leverage.
The risks
These are among the most volatile equities on earth. The multiples are stretched, the floats are tiny, and any broad AI sentiment shift hits names like these first and hardest. More immediately: cornerstone lock-ups on both expire around July 8th, meaning a meaningful chunk of new float hits the market right when retail attention is building. If you're trading these, size carefully and watch your leverage.